1 /* $NetBSD: bt_open.c,v 1.10 1997/07/21 14:06:32 jtc 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 1.10 1997/07/21 14:06:32 jtc 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 69 #include <db.h> 70 #include "btree.h" 71 72 #ifdef DEBUG 73 #undef MINPSIZE 74 #define MINPSIZE 128 75 #endif 76 77 static int byteorder __P((void)); 78 static int nroot __P((BTREE *)); 79 static int tmp __P((void)); 80 81 /* 82 * __BT_OPEN -- Open a btree. 83 * 84 * Creates and fills a DB struct, and calls the routine that actually 85 * opens the btree. 86 * 87 * Parameters: 88 * fname: filename (NULL for in-memory trees) 89 * flags: open flag bits 90 * mode: open permission bits 91 * b: BTREEINFO pointer 92 * 93 * Returns: 94 * NULL on failure, pointer to DB on success. 95 * 96 */ 97 DB * 98 __bt_open(fname, flags, mode, openinfo, dflags) 99 const char *fname; 100 int flags, mode, dflags; 101 const BTREEINFO *openinfo; 102 { 103 struct stat sb; 104 BTMETA m; 105 BTREE *t; 106 BTREEINFO b; 107 DB *dbp; 108 pgno_t ncache; 109 ssize_t nr; 110 int machine_lorder; 111 112 t = NULL; 113 114 /* 115 * Intention is to make sure all of the user's selections are okay 116 * here and then use them without checking. Can't be complete, since 117 * we don't know the right page size, lorder or flags until the backing 118 * file is opened. Also, the file's page size can cause the cachesize 119 * to change. 120 */ 121 machine_lorder = byteorder(); 122 if (openinfo) { 123 b = *openinfo; 124 125 /* Flags: R_DUP. */ 126 if (b.flags & ~(R_DUP)) 127 goto einval; 128 129 /* 130 * Page size must be indx_t aligned and >= MINPSIZE. Default 131 * page size is set farther on, based on the underlying file 132 * transfer size. 133 */ 134 if (b.psize && 135 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 136 b.psize & (sizeof(indx_t) - 1))) 137 goto einval; 138 139 /* Minimum number of keys per page; absolute minimum is 2. */ 140 if (b.minkeypage) { 141 if (b.minkeypage < 2) 142 goto einval; 143 } else 144 b.minkeypage = DEFMINKEYPAGE; 145 146 /* If no comparison, use default comparison and prefix. */ 147 if (b.compare == NULL) { 148 b.compare = __bt_defcmp; 149 if (b.prefix == NULL) 150 b.prefix = __bt_defpfx; 151 } 152 153 if (b.lorder == 0) 154 b.lorder = machine_lorder; 155 } else { 156 b.compare = __bt_defcmp; 157 b.cachesize = 0; 158 b.flags = 0; 159 b.lorder = machine_lorder; 160 b.minkeypage = DEFMINKEYPAGE; 161 b.prefix = __bt_defpfx; 162 b.psize = 0; 163 } 164 165 /* Check for the ubiquitous PDP-11. */ 166 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 167 goto einval; 168 169 /* Allocate and initialize DB and BTREE structures. */ 170 if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL) 171 goto err; 172 memset(t, 0, sizeof(BTREE)); 173 t->bt_fd = -1; /* Don't close unopened fd on error. */ 174 t->bt_lorder = b.lorder; 175 t->bt_order = NOT; 176 t->bt_cmp = b.compare; 177 t->bt_pfx = b.prefix; 178 t->bt_rfd = -1; 179 180 if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL) 181 goto err; 182 memset(t->bt_dbp, 0, sizeof(DB)); 183 if (t->bt_lorder != machine_lorder) 184 F_SET(t, B_NEEDSWAP); 185 186 dbp->type = DB_BTREE; 187 dbp->internal = t; 188 dbp->close = __bt_close; 189 dbp->del = __bt_delete; 190 dbp->fd = __bt_fd; 191 dbp->get = __bt_get; 192 dbp->put = __bt_put; 193 dbp->seq = __bt_seq; 194 dbp->sync = __bt_sync; 195 196 /* 197 * If no file name was supplied, this is an in-memory btree and we 198 * open a backing temporary file. Otherwise, it's a disk-based tree. 199 */ 200 if (fname) { 201 switch (flags & O_ACCMODE) { 202 case O_RDONLY: 203 F_SET(t, B_RDONLY); 204 break; 205 case O_RDWR: 206 break; 207 case O_WRONLY: 208 default: 209 goto einval; 210 } 211 212 if ((t->bt_fd = open(fname, flags, mode)) < 0) 213 goto err; 214 215 } else { 216 if ((flags & O_ACCMODE) != O_RDWR) 217 goto einval; 218 if ((t->bt_fd = tmp()) == -1) 219 goto err; 220 F_SET(t, B_INMEM); 221 } 222 223 if (fcntl(t->bt_fd, F_SETFD, 1) == -1) 224 goto err; 225 226 if (fstat(t->bt_fd, &sb)) 227 goto err; 228 if (sb.st_size) { 229 if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 230 goto err; 231 if (nr != sizeof(BTMETA)) 232 goto eftype; 233 234 /* 235 * Read in the meta-data. This can change the notion of what 236 * the lorder, page size and flags are, and, when the page size 237 * changes, the cachesize value can change too. If the user 238 * specified the wrong byte order for an existing database, we 239 * don't bother to return an error, we just clear the NEEDSWAP 240 * bit. 241 */ 242 if (m.magic == BTREEMAGIC) 243 F_CLR(t, B_NEEDSWAP); 244 else { 245 F_SET(t, B_NEEDSWAP); 246 M_32_SWAP(m.magic); 247 M_32_SWAP(m.version); 248 M_32_SWAP(m.psize); 249 M_32_SWAP(m.free); 250 M_32_SWAP(m.nrecs); 251 M_32_SWAP(m.flags); 252 } 253 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 254 goto eftype; 255 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 256 m.psize & (sizeof(indx_t) - 1)) 257 goto eftype; 258 if (m.flags & ~SAVEMETA) 259 goto eftype; 260 b.psize = m.psize; 261 F_SET(t, m.flags); 262 t->bt_free = m.free; 263 t->bt_nrecs = m.nrecs; 264 } else { 265 /* 266 * Set the page size to the best value for I/O to this file. 267 * Don't overflow the page offset type. 268 */ 269 if (b.psize == 0) { 270 b.psize = sb.st_blksize; 271 if (b.psize < MINPSIZE) 272 b.psize = MINPSIZE; 273 if (b.psize > MAX_PAGE_OFFSET + 1) 274 b.psize = MAX_PAGE_OFFSET + 1; 275 } 276 277 /* Set flag if duplicates permitted. */ 278 if (!(b.flags & R_DUP)) 279 F_SET(t, B_NODUPS); 280 281 t->bt_free = P_INVALID; 282 t->bt_nrecs = 0; 283 F_SET(t, B_METADIRTY); 284 } 285 286 t->bt_psize = b.psize; 287 288 /* Set the cache size; must be a multiple of the page size. */ 289 if (b.cachesize && b.cachesize & (b.psize - 1)) 290 b.cachesize += (~b.cachesize & (b.psize - 1)) + 1; 291 if (b.cachesize < b.psize * MINCACHE) 292 b.cachesize = b.psize * MINCACHE; 293 294 /* Calculate number of pages to cache. */ 295 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 296 297 /* 298 * The btree data structure requires that at least two keys can fit on 299 * a page, but other than that there's no fixed requirement. The user 300 * specified a minimum number per page, and we translated that into the 301 * number of bytes a key/data pair can use before being placed on an 302 * overflow page. This calculation includes the page header, the size 303 * of the index referencing the leaf item and the size of the leaf item 304 * structure. Also, don't let the user specify a minkeypage such that 305 * a key/data pair won't fit even if both key and data are on overflow 306 * pages. 307 */ 308 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 309 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 310 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 311 t->bt_ovflsize = 312 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 313 314 /* Initialize the buffer pool. */ 315 if ((t->bt_mp = 316 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 317 goto err; 318 if (!F_ISSET(t, B_INMEM)) 319 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 320 321 /* Create a root page if new tree. */ 322 if (nroot(t) == RET_ERROR) 323 goto err; 324 325 /* Global flags. */ 326 if (dflags & DB_LOCK) 327 F_SET(t, B_DB_LOCK); 328 if (dflags & DB_SHMEM) 329 F_SET(t, B_DB_SHMEM); 330 if (dflags & DB_TXN) 331 F_SET(t, B_DB_TXN); 332 333 return (dbp); 334 335 einval: errno = EINVAL; 336 goto err; 337 338 eftype: errno = EFTYPE; 339 goto err; 340 341 err: if (t) { 342 if (t->bt_dbp) 343 free(t->bt_dbp); 344 if (t->bt_fd != -1) 345 (void)close(t->bt_fd); 346 free(t); 347 } 348 return (NULL); 349 } 350 351 /* 352 * NROOT -- Create the root of a new tree. 353 * 354 * Parameters: 355 * t: tree 356 * 357 * Returns: 358 * RET_ERROR, RET_SUCCESS 359 */ 360 static int 361 nroot(t) 362 BTREE *t; 363 { 364 PAGE *meta, *root; 365 pgno_t npg; 366 367 if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { 368 mpool_put(t->bt_mp, meta, 0); 369 return (RET_SUCCESS); 370 } 371 if (errno != EINVAL) /* It's OK to not exist. */ 372 return (RET_ERROR); 373 errno = 0; 374 375 if ((meta = mpool_new(t->bt_mp, &npg)) == NULL) 376 return (RET_ERROR); 377 378 if ((root = mpool_new(t->bt_mp, &npg)) == NULL) 379 return (RET_ERROR); 380 381 if (npg != P_ROOT) 382 return (RET_ERROR); 383 root->pgno = npg; 384 root->prevpg = root->nextpg = P_INVALID; 385 root->lower = BTDATAOFF; 386 root->upper = t->bt_psize; 387 root->flags = P_BLEAF; 388 memset(meta, 0, t->bt_psize); 389 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 390 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 391 return (RET_SUCCESS); 392 } 393 394 static int 395 tmp() 396 { 397 sigset_t set, oset; 398 int fd; 399 char *envtmp; 400 char path[MAXPATHLEN]; 401 402 envtmp = getenv("TMPDIR"); 403 (void)snprintf(path, 404 sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : "/tmp"); 405 406 (void)sigfillset(&set); 407 (void)sigprocmask(SIG_BLOCK, &set, &oset); 408 if ((fd = mkstemp(path)) != -1) 409 (void)unlink(path); 410 (void)sigprocmask(SIG_SETMASK, &oset, NULL); 411 return(fd); 412 } 413 414 static int 415 byteorder() 416 { 417 u_int32_t x; 418 u_char *p; 419 420 x = 0x01020304; 421 p = (u_char *)&x; 422 switch (*p) { 423 case 1: 424 return (BIG_ENDIAN); 425 case 4: 426 return (LITTLE_ENDIAN); 427 default: 428 return (0); 429 } 430 } 431 432 int 433 __bt_fd(dbp) 434 const DB *dbp; 435 { 436 BTREE *t; 437 438 t = dbp->internal; 439 440 /* Toss any page pinned across calls. */ 441 if (t->bt_pinned != NULL) { 442 mpool_put(t->bt_mp, t->bt_pinned, 0); 443 t->bt_pinned = NULL; 444 } 445 446 /* In-memory database can't have a file descriptor. */ 447 if (F_ISSET(t, B_INMEM)) { 448 errno = ENOENT; 449 return (-1); 450 } 451 return (t->bt_fd); 452 } 453