1 /* $NetBSD: bt_open.c,v 1.11 1998/04/07 10:40:21 fair 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. 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 #include <sys/cdefs.h> 40 #if defined(LIBC_SCCS) && !defined(lint) 41 #if 0 42 static char sccsid[] = "@(#)bt_open.c 8.10 (Berkeley) 8/17/94"; 43 #else 44 __RCSID("$NetBSD: bt_open.c,v 1.11 1998/04/07 10:40:21 fair Exp $"); 45 #endif 46 #endif /* LIBC_SCCS and not lint */ 47 48 /* 49 * Implementation of btree access method for 4.4BSD. 50 * 51 * The design here was originally based on that of the btree access method 52 * used in the Postgres database system at UC Berkeley. This implementation 53 * is wholly independent of the Postgres code. 54 */ 55 56 #include "namespace.h" 57 #include <sys/param.h> 58 #include <sys/stat.h> 59 60 #include <errno.h> 61 #include <fcntl.h> 62 #include <limits.h> 63 #include <signal.h> 64 #include <stdio.h> 65 #include <stdlib.h> 66 #include <string.h> 67 #include <unistd.h> 68 #include <paths.h> 69 70 #include <db.h> 71 #include "btree.h" 72 73 #ifdef DEBUG 74 #undef MINPSIZE 75 #define MINPSIZE 128 76 #endif 77 78 static int byteorder __P((void)); 79 static int nroot __P((BTREE *)); 80 static int tmp __P((void)); 81 82 /* 83 * __BT_OPEN -- Open a btree. 84 * 85 * Creates and fills a DB struct, and calls the routine that actually 86 * opens the btree. 87 * 88 * Parameters: 89 * fname: filename (NULL for in-memory trees) 90 * flags: open flag bits 91 * mode: open permission bits 92 * b: BTREEINFO pointer 93 * 94 * Returns: 95 * NULL on failure, pointer to DB on success. 96 * 97 */ 98 DB * 99 __bt_open(fname, flags, mode, openinfo, dflags) 100 const char *fname; 101 int flags, mode, dflags; 102 const BTREEINFO *openinfo; 103 { 104 struct stat sb; 105 BTMETA m; 106 BTREE *t; 107 BTREEINFO b; 108 DB *dbp; 109 pgno_t ncache; 110 ssize_t nr; 111 int machine_lorder; 112 113 t = NULL; 114 115 /* 116 * Intention is to make sure all of the user's selections are okay 117 * here and then use them without checking. Can't be complete, since 118 * we don't know the right page size, lorder or flags until the backing 119 * file is opened. Also, the file's page size can cause the cachesize 120 * to change. 121 */ 122 machine_lorder = byteorder(); 123 if (openinfo) { 124 b = *openinfo; 125 126 /* Flags: R_DUP. */ 127 if (b.flags & ~(R_DUP)) 128 goto einval; 129 130 /* 131 * Page size must be indx_t aligned and >= MINPSIZE. Default 132 * page size is set farther on, based on the underlying file 133 * transfer size. 134 */ 135 if (b.psize && 136 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 137 b.psize & (sizeof(indx_t) - 1))) 138 goto einval; 139 140 /* Minimum number of keys per page; absolute minimum is 2. */ 141 if (b.minkeypage) { 142 if (b.minkeypage < 2) 143 goto einval; 144 } else 145 b.minkeypage = DEFMINKEYPAGE; 146 147 /* If no comparison, use default comparison and prefix. */ 148 if (b.compare == NULL) { 149 b.compare = __bt_defcmp; 150 if (b.prefix == NULL) 151 b.prefix = __bt_defpfx; 152 } 153 154 if (b.lorder == 0) 155 b.lorder = machine_lorder; 156 } else { 157 b.compare = __bt_defcmp; 158 b.cachesize = 0; 159 b.flags = 0; 160 b.lorder = machine_lorder; 161 b.minkeypage = DEFMINKEYPAGE; 162 b.prefix = __bt_defpfx; 163 b.psize = 0; 164 } 165 166 /* Check for the ubiquitous PDP-11. */ 167 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 168 goto einval; 169 170 /* Allocate and initialize DB and BTREE structures. */ 171 if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL) 172 goto err; 173 memset(t, 0, sizeof(BTREE)); 174 t->bt_fd = -1; /* Don't close unopened fd on error. */ 175 t->bt_lorder = b.lorder; 176 t->bt_order = NOT; 177 t->bt_cmp = b.compare; 178 t->bt_pfx = b.prefix; 179 t->bt_rfd = -1; 180 181 if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL) 182 goto err; 183 memset(t->bt_dbp, 0, sizeof(DB)); 184 if (t->bt_lorder != machine_lorder) 185 F_SET(t, B_NEEDSWAP); 186 187 dbp->type = DB_BTREE; 188 dbp->internal = t; 189 dbp->close = __bt_close; 190 dbp->del = __bt_delete; 191 dbp->fd = __bt_fd; 192 dbp->get = __bt_get; 193 dbp->put = __bt_put; 194 dbp->seq = __bt_seq; 195 dbp->sync = __bt_sync; 196 197 /* 198 * If no file name was supplied, this is an in-memory btree and we 199 * open a backing temporary file. Otherwise, it's a disk-based tree. 200 */ 201 if (fname) { 202 switch (flags & O_ACCMODE) { 203 case O_RDONLY: 204 F_SET(t, B_RDONLY); 205 break; 206 case O_RDWR: 207 break; 208 case O_WRONLY: 209 default: 210 goto einval; 211 } 212 213 if ((t->bt_fd = open(fname, flags, mode)) < 0) 214 goto err; 215 216 } else { 217 if ((flags & O_ACCMODE) != O_RDWR) 218 goto einval; 219 if ((t->bt_fd = tmp()) == -1) 220 goto err; 221 F_SET(t, B_INMEM); 222 } 223 224 if (fcntl(t->bt_fd, F_SETFD, 1) == -1) 225 goto err; 226 227 if (fstat(t->bt_fd, &sb)) 228 goto err; 229 if (sb.st_size) { 230 if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 231 goto err; 232 if (nr != sizeof(BTMETA)) 233 goto eftype; 234 235 /* 236 * Read in the meta-data. This can change the notion of what 237 * the lorder, page size and flags are, and, when the page size 238 * changes, the cachesize value can change too. If the user 239 * specified the wrong byte order for an existing database, we 240 * don't bother to return an error, we just clear the NEEDSWAP 241 * bit. 242 */ 243 if (m.magic == BTREEMAGIC) 244 F_CLR(t, B_NEEDSWAP); 245 else { 246 F_SET(t, B_NEEDSWAP); 247 M_32_SWAP(m.magic); 248 M_32_SWAP(m.version); 249 M_32_SWAP(m.psize); 250 M_32_SWAP(m.free); 251 M_32_SWAP(m.nrecs); 252 M_32_SWAP(m.flags); 253 } 254 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 255 goto eftype; 256 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 257 m.psize & (sizeof(indx_t) - 1)) 258 goto eftype; 259 if (m.flags & ~SAVEMETA) 260 goto eftype; 261 b.psize = m.psize; 262 F_SET(t, m.flags); 263 t->bt_free = m.free; 264 t->bt_nrecs = m.nrecs; 265 } else { 266 /* 267 * Set the page size to the best value for I/O to this file. 268 * Don't overflow the page offset type. 269 */ 270 if (b.psize == 0) { 271 b.psize = sb.st_blksize; 272 if (b.psize < MINPSIZE) 273 b.psize = MINPSIZE; 274 if (b.psize > MAX_PAGE_OFFSET + 1) 275 b.psize = MAX_PAGE_OFFSET + 1; 276 } 277 278 /* Set flag if duplicates permitted. */ 279 if (!(b.flags & R_DUP)) 280 F_SET(t, B_NODUPS); 281 282 t->bt_free = P_INVALID; 283 t->bt_nrecs = 0; 284 F_SET(t, B_METADIRTY); 285 } 286 287 t->bt_psize = b.psize; 288 289 /* Set the cache size; must be a multiple of the page size. */ 290 if (b.cachesize && b.cachesize & (b.psize - 1)) 291 b.cachesize += (~b.cachesize & (b.psize - 1)) + 1; 292 if (b.cachesize < b.psize * MINCACHE) 293 b.cachesize = b.psize * MINCACHE; 294 295 /* Calculate number of pages to cache. */ 296 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 297 298 /* 299 * The btree data structure requires that at least two keys can fit on 300 * a page, but other than that there's no fixed requirement. The user 301 * specified a minimum number per page, and we translated that into the 302 * number of bytes a key/data pair can use before being placed on an 303 * overflow page. This calculation includes the page header, the size 304 * of the index referencing the leaf item and the size of the leaf item 305 * structure. Also, don't let the user specify a minkeypage such that 306 * a key/data pair won't fit even if both key and data are on overflow 307 * pages. 308 */ 309 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 310 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 311 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 312 t->bt_ovflsize = 313 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 314 315 /* Initialize the buffer pool. */ 316 if ((t->bt_mp = 317 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 318 goto err; 319 if (!F_ISSET(t, B_INMEM)) 320 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 321 322 /* Create a root page if new tree. */ 323 if (nroot(t) == RET_ERROR) 324 goto err; 325 326 /* Global flags. */ 327 if (dflags & DB_LOCK) 328 F_SET(t, B_DB_LOCK); 329 if (dflags & DB_SHMEM) 330 F_SET(t, B_DB_SHMEM); 331 if (dflags & DB_TXN) 332 F_SET(t, B_DB_TXN); 333 334 return (dbp); 335 336 einval: errno = EINVAL; 337 goto err; 338 339 eftype: errno = EFTYPE; 340 goto err; 341 342 err: if (t) { 343 if (t->bt_dbp) 344 free(t->bt_dbp); 345 if (t->bt_fd != -1) 346 (void)close(t->bt_fd); 347 free(t); 348 } 349 return (NULL); 350 } 351 352 /* 353 * NROOT -- Create the root of a new tree. 354 * 355 * Parameters: 356 * t: tree 357 * 358 * Returns: 359 * RET_ERROR, RET_SUCCESS 360 */ 361 static int 362 nroot(t) 363 BTREE *t; 364 { 365 PAGE *meta, *root; 366 pgno_t npg; 367 368 if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { 369 mpool_put(t->bt_mp, meta, 0); 370 return (RET_SUCCESS); 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)) == NULL) 377 return (RET_ERROR); 378 379 if ((root = mpool_new(t->bt_mp, &npg)) == 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 sigset_t set, oset; 399 int fd; 400 char *envtmp; 401 char path[MAXPATHLEN]; 402 403 envtmp = getenv("TMPDIR"); 404 (void)snprintf(path, 405 sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : _PATH_TMP); 406 407 (void)sigfillset(&set); 408 (void)sigprocmask(SIG_BLOCK, &set, &oset); 409 if ((fd = mkstemp(path)) != -1) 410 (void)unlink(path); 411 (void)sigprocmask(SIG_SETMASK, &oset, NULL); 412 return(fd); 413 } 414 415 static int 416 byteorder() 417 { 418 u_int32_t x; 419 u_char *p; 420 421 x = 0x01020304; 422 p = (u_char *)&x; 423 switch (*p) { 424 case 1: 425 return (BIG_ENDIAN); 426 case 4: 427 return (LITTLE_ENDIAN); 428 default: 429 return (0); 430 } 431 } 432 433 int 434 __bt_fd(dbp) 435 const DB *dbp; 436 { 437 BTREE *t; 438 439 t = dbp->internal; 440 441 /* Toss any page pinned across calls. */ 442 if (t->bt_pinned != NULL) { 443 mpool_put(t->bt_mp, t->bt_pinned, 0); 444 t->bt_pinned = NULL; 445 } 446 447 /* In-memory database can't have a file descriptor. */ 448 if (F_ISSET(t, B_INMEM)) { 449 errno = ENOENT; 450 return (-1); 451 } 452 return (t->bt_fd); 453 } 454