1 /* $NetBSD: bt_open.c,v 1.8 1996/05/03 21:50:46 cgd 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 #if defined(LIBC_SCCS) && !defined(lint) 40 #if 0 41 static char sccsid[] = "@(#)bt_open.c 8.10 (Berkeley) 8/17/94"; 42 #else 43 static char rcsid[] = "$NetBSD: bt_open.c,v 1.8 1996/05/03 21:50:46 cgd Exp $"; 44 #endif 45 #endif /* LIBC_SCCS and not lint */ 46 47 /* 48 * Implementation of btree access method for 4.4BSD. 49 * 50 * The design here was originally based on that of the btree access method 51 * used in the Postgres database system at UC Berkeley. This implementation 52 * is wholly independent of the Postgres code. 53 */ 54 55 #include <sys/param.h> 56 #include <sys/stat.h> 57 58 #include <errno.h> 59 #include <fcntl.h> 60 #include <limits.h> 61 #include <signal.h> 62 #include <stdio.h> 63 #include <stdlib.h> 64 #include <string.h> 65 #include <unistd.h> 66 67 #include <db.h> 68 #include "btree.h" 69 70 #ifdef DEBUG 71 #undef MINPSIZE 72 #define MINPSIZE 128 73 #endif 74 75 static int byteorder __P((void)); 76 static int nroot __P((BTREE *)); 77 static int tmp __P((void)); 78 79 /* 80 * __BT_OPEN -- Open a btree. 81 * 82 * Creates and fills a DB struct, and calls the routine that actually 83 * opens the btree. 84 * 85 * Parameters: 86 * fname: filename (NULL for in-memory trees) 87 * flags: open flag bits 88 * mode: open permission bits 89 * b: BTREEINFO pointer 90 * 91 * Returns: 92 * NULL on failure, pointer to DB on success. 93 * 94 */ 95 DB * 96 __bt_open(fname, flags, mode, openinfo, dflags) 97 const char *fname; 98 int flags, mode, dflags; 99 const BTREEINFO *openinfo; 100 { 101 struct stat sb; 102 BTMETA m; 103 BTREE *t; 104 BTREEINFO b; 105 DB *dbp; 106 pgno_t ncache; 107 ssize_t nr; 108 int machine_lorder; 109 110 t = NULL; 111 112 /* 113 * Intention is to make sure all of the user's selections are okay 114 * here and then use them without checking. Can't be complete, since 115 * we don't know the right page size, lorder or flags until the backing 116 * file is opened. Also, the file's page size can cause the cachesize 117 * to change. 118 */ 119 machine_lorder = byteorder(); 120 if (openinfo) { 121 b = *openinfo; 122 123 /* Flags: R_DUP. */ 124 if (b.flags & ~(R_DUP)) 125 goto einval; 126 127 /* 128 * Page size must be indx_t aligned and >= MINPSIZE. Default 129 * page size is set farther on, based on the underlying file 130 * transfer size. 131 */ 132 if (b.psize && 133 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 134 b.psize & sizeof(indx_t) - 1)) 135 goto einval; 136 137 /* Minimum number of keys per page; absolute minimum is 2. */ 138 if (b.minkeypage) { 139 if (b.minkeypage < 2) 140 goto einval; 141 } else 142 b.minkeypage = DEFMINKEYPAGE; 143 144 /* If no comparison, use default comparison and prefix. */ 145 if (b.compare == NULL) { 146 b.compare = __bt_defcmp; 147 if (b.prefix == NULL) 148 b.prefix = __bt_defpfx; 149 } 150 151 if (b.lorder == 0) 152 b.lorder = machine_lorder; 153 } else { 154 b.compare = __bt_defcmp; 155 b.cachesize = 0; 156 b.flags = 0; 157 b.lorder = machine_lorder; 158 b.minkeypage = DEFMINKEYPAGE; 159 b.prefix = __bt_defpfx; 160 b.psize = 0; 161 } 162 163 /* Check for the ubiquitous PDP-11. */ 164 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 165 goto einval; 166 167 /* Allocate and initialize DB and BTREE structures. */ 168 if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL) 169 goto err; 170 memset(t, 0, sizeof(BTREE)); 171 t->bt_fd = -1; /* Don't close unopened fd on error. */ 172 t->bt_lorder = b.lorder; 173 t->bt_order = NOT; 174 t->bt_cmp = b.compare; 175 t->bt_pfx = b.prefix; 176 t->bt_rfd = -1; 177 178 if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL) 179 goto err; 180 memset(t->bt_dbp, 0, sizeof(DB)); 181 if (t->bt_lorder != machine_lorder) 182 F_SET(t, B_NEEDSWAP); 183 184 dbp->type = DB_BTREE; 185 dbp->internal = t; 186 dbp->close = __bt_close; 187 dbp->del = __bt_delete; 188 dbp->fd = __bt_fd; 189 dbp->get = __bt_get; 190 dbp->put = __bt_put; 191 dbp->seq = __bt_seq; 192 dbp->sync = __bt_sync; 193 194 /* 195 * If no file name was supplied, this is an in-memory btree and we 196 * open a backing temporary file. Otherwise, it's a disk-based tree. 197 */ 198 if (fname) { 199 switch (flags & O_ACCMODE) { 200 case O_RDONLY: 201 F_SET(t, B_RDONLY); 202 break; 203 case O_RDWR: 204 break; 205 case O_WRONLY: 206 default: 207 goto einval; 208 } 209 210 if ((t->bt_fd = open(fname, flags, mode)) < 0) 211 goto err; 212 213 } else { 214 if ((flags & O_ACCMODE) != O_RDWR) 215 goto einval; 216 if ((t->bt_fd = tmp()) == -1) 217 goto err; 218 F_SET(t, B_INMEM); 219 } 220 221 if (fcntl(t->bt_fd, F_SETFD, 1) == -1) 222 goto err; 223 224 if (fstat(t->bt_fd, &sb)) 225 goto err; 226 if (sb.st_size) { 227 if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 228 goto err; 229 if (nr != sizeof(BTMETA)) 230 goto eftype; 231 232 /* 233 * Read in the meta-data. This can change the notion of what 234 * the lorder, page size and flags are, and, when the page size 235 * changes, the cachesize value can change too. If the user 236 * specified the wrong byte order for an existing database, we 237 * don't bother to return an error, we just clear the NEEDSWAP 238 * bit. 239 */ 240 if (m.magic == BTREEMAGIC) 241 F_CLR(t, B_NEEDSWAP); 242 else { 243 F_SET(t, B_NEEDSWAP); 244 M_32_SWAP(m.magic); 245 M_32_SWAP(m.version); 246 M_32_SWAP(m.psize); 247 M_32_SWAP(m.free); 248 M_32_SWAP(m.nrecs); 249 M_32_SWAP(m.flags); 250 } 251 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 252 goto eftype; 253 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 254 m.psize & sizeof(indx_t) - 1) 255 goto eftype; 256 if (m.flags & ~SAVEMETA) 257 goto eftype; 258 b.psize = m.psize; 259 F_SET(t, m.flags); 260 t->bt_free = m.free; 261 t->bt_nrecs = m.nrecs; 262 } else { 263 /* 264 * Set the page size to the best value for I/O to this file. 265 * Don't overflow the page offset type. 266 */ 267 if (b.psize == 0) { 268 b.psize = sb.st_blksize; 269 if (b.psize < MINPSIZE) 270 b.psize = MINPSIZE; 271 if (b.psize > MAX_PAGE_OFFSET + 1) 272 b.psize = MAX_PAGE_OFFSET + 1; 273 } 274 275 /* Set flag if duplicates permitted. */ 276 if (!(b.flags & R_DUP)) 277 F_SET(t, B_NODUPS); 278 279 t->bt_free = P_INVALID; 280 t->bt_nrecs = 0; 281 F_SET(t, B_METADIRTY); 282 } 283 284 t->bt_psize = b.psize; 285 286 /* Set the cache size; must be a multiple of the page size. */ 287 if (b.cachesize && b.cachesize & b.psize - 1) 288 b.cachesize += (~b.cachesize & b.psize - 1) + 1; 289 if (b.cachesize < b.psize * MINCACHE) 290 b.cachesize = b.psize * MINCACHE; 291 292 /* Calculate number of pages to cache. */ 293 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 294 295 /* 296 * The btree data structure requires that at least two keys can fit on 297 * a page, but other than that there's no fixed requirement. The user 298 * specified a minimum number per page, and we translated that into the 299 * number of bytes a key/data pair can use before being placed on an 300 * overflow page. This calculation includes the page header, the size 301 * of the index referencing the leaf item and the size of the leaf item 302 * structure. Also, don't let the user specify a minkeypage such that 303 * a key/data pair won't fit even if both key and data are on overflow 304 * pages. 305 */ 306 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 307 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 308 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 309 t->bt_ovflsize = 310 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 311 312 /* Initialize the buffer pool. */ 313 if ((t->bt_mp = 314 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 315 goto err; 316 if (!F_ISSET(t, B_INMEM)) 317 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 318 319 /* Create a root page if new tree. */ 320 if (nroot(t) == RET_ERROR) 321 goto err; 322 323 /* Global flags. */ 324 if (dflags & DB_LOCK) 325 F_SET(t, B_DB_LOCK); 326 if (dflags & DB_SHMEM) 327 F_SET(t, B_DB_SHMEM); 328 if (dflags & DB_TXN) 329 F_SET(t, B_DB_TXN); 330 331 return (dbp); 332 333 einval: errno = EINVAL; 334 goto err; 335 336 eftype: errno = EFTYPE; 337 goto err; 338 339 err: if (t) { 340 if (t->bt_dbp) 341 free(t->bt_dbp); 342 if (t->bt_fd != -1) 343 (void)close(t->bt_fd); 344 free(t); 345 } 346 return (NULL); 347 } 348 349 /* 350 * NROOT -- Create the root of a new tree. 351 * 352 * Parameters: 353 * t: tree 354 * 355 * Returns: 356 * RET_ERROR, RET_SUCCESS 357 */ 358 static int 359 nroot(t) 360 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() 394 { 395 sigset_t set, oset; 396 int fd; 397 char *envtmp; 398 char path[MAXPATHLEN]; 399 400 envtmp = getenv("TMPDIR"); 401 (void)snprintf(path, 402 sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : "/tmp"); 403 404 (void)sigfillset(&set); 405 (void)sigprocmask(SIG_BLOCK, &set, &oset); 406 if ((fd = mkstemp(path)) != -1) 407 (void)unlink(path); 408 (void)sigprocmask(SIG_SETMASK, &oset, NULL); 409 return(fd); 410 } 411 412 static int 413 byteorder() 414 { 415 u_int32_t x; 416 u_char *p; 417 418 x = 0x01020304; 419 p = (u_char *)&x; 420 switch (*p) { 421 case 1: 422 return (BIG_ENDIAN); 423 case 4: 424 return (LITTLE_ENDIAN); 425 default: 426 return (0); 427 } 428 } 429 430 int 431 __bt_fd(dbp) 432 const DB *dbp; 433 { 434 BTREE *t; 435 436 t = dbp->internal; 437 438 /* Toss any page pinned across calls. */ 439 if (t->bt_pinned != NULL) { 440 mpool_put(t->bt_mp, t->bt_pinned, 0); 441 t->bt_pinned = NULL; 442 } 443 444 /* In-memory database can't have a file descriptor. */ 445 if (F_ISSET(t, B_INMEM)) { 446 errno = ENOENT; 447 return (-1); 448 } 449 return (t->bt_fd); 450 } 451