1 /* $NetBSD: bt_open.c,v 1.18 2005/01/19 00:23:44 mycroft 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 #include <sys/cdefs.h> 36 #if defined(LIBC_SCCS) && !defined(lint) 37 #if 0 38 static char sccsid[] = "@(#)bt_open.c 8.10 (Berkeley) 8/17/94"; 39 #else 40 __RCSID("$NetBSD: bt_open.c,v 1.18 2005/01/19 00:23:44 mycroft Exp $"); 41 #endif 42 #endif /* LIBC_SCCS and not lint */ 43 44 /* 45 * Implementation of btree access method for 4.4BSD. 46 * 47 * The design here was originally based on that of the btree access method 48 * used in the Postgres database system at UC Berkeley. This implementation 49 * is wholly independent of the Postgres code. 50 */ 51 52 #include "namespace.h" 53 #include <sys/stat.h> 54 55 #include <errno.h> 56 #include <fcntl.h> 57 #include <limits.h> 58 #include <signal.h> 59 #include <stdio.h> 60 #include <stdlib.h> 61 #include <string.h> 62 #include <unistd.h> 63 #include <paths.h> 64 65 #include <db.h> 66 #include "btree.h" 67 68 #ifdef DEBUG 69 #undef MINPSIZE 70 #define MINPSIZE 128 71 #endif 72 73 static int byteorder __P((void)); 74 static int nroot __P((BTREE *)); 75 static int tmp __P((void)); 76 77 /* 78 * __BT_OPEN -- Open a btree. 79 * 80 * Creates and fills a DB struct, and calls the routine that actually 81 * opens the btree. 82 * 83 * Parameters: 84 * fname: filename (NULL for in-memory trees) 85 * flags: open flag bits 86 * mode: open permission bits 87 * b: BTREEINFO pointer 88 * 89 * Returns: 90 * NULL on failure, pointer to DB on success. 91 * 92 */ 93 DB * 94 __bt_open(fname, flags, mode, openinfo, dflags) 95 const char *fname; 96 int flags; 97 mode_t mode; 98 const BTREEINFO *openinfo; 99 int dflags; 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)) == -1) 211 goto err; 212 if (fcntl(t->bt_fd, F_SETFD, FD_CLOEXEC) == -1) 213 goto err; 214 } else { 215 if ((flags & O_ACCMODE) != O_RDWR) 216 goto einval; 217 if ((t->bt_fd = tmp()) == -1) 218 goto err; 219 F_SET(t, B_INMEM); 220 } 221 222 if (fcntl(t->bt_fd, F_SETFD, 1) == -1) 223 goto err; 224 225 if (fstat(t->bt_fd, &sb)) 226 goto err; 227 if (sb.st_size) { 228 if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 229 goto err; 230 if (nr != sizeof(BTMETA)) 231 goto eftype; 232 233 /* 234 * Read in the meta-data. This can change the notion of what 235 * the lorder, page size and flags are, and, when the page size 236 * changes, the cachesize value can change too. If the user 237 * specified the wrong byte order for an existing database, we 238 * don't bother to return an error, we just clear the NEEDSWAP 239 * bit. 240 */ 241 if (m.magic == BTREEMAGIC) 242 F_CLR(t, B_NEEDSWAP); 243 else { 244 F_SET(t, B_NEEDSWAP); 245 M_32_SWAP(m.magic); 246 M_32_SWAP(m.version); 247 M_32_SWAP(m.psize); 248 M_32_SWAP(m.free); 249 M_32_SWAP(m.nrecs); 250 M_32_SWAP(m.flags); 251 } 252 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 253 goto eftype; 254 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 255 m.psize & (sizeof(indx_t) - 1)) 256 goto eftype; 257 if (m.flags & ~SAVEMETA) 258 goto eftype; 259 b.psize = m.psize; 260 F_SET(t, m.flags); 261 t->bt_free = m.free; 262 t->bt_nrecs = m.nrecs; 263 } else { 264 /* 265 * Set the page size to the best value for I/O to this file. 266 * Don't overflow the page offset type. 267 */ 268 if (b.psize == 0) { 269 b.psize = sb.st_blksize; 270 if (b.psize < MINPSIZE) 271 b.psize = MINPSIZE; 272 if (b.psize > MAX_PAGE_OFFSET + 1) 273 b.psize = MAX_PAGE_OFFSET + 1; 274 } 275 276 /* Set flag if duplicates permitted. */ 277 if (!(b.flags & R_DUP)) 278 F_SET(t, B_NODUPS); 279 280 t->bt_free = P_INVALID; 281 t->bt_nrecs = 0; 282 F_SET(t, B_METADIRTY); 283 } 284 285 t->bt_psize = b.psize; 286 287 /* Set the cache size; must be a multiple of the page size. */ 288 if (b.cachesize && b.cachesize & (b.psize - 1)) 289 b.cachesize += (~b.cachesize & (b.psize - 1)) + 1; 290 if (b.cachesize < b.psize * MINCACHE) 291 b.cachesize = b.psize * MINCACHE; 292 293 /* Calculate number of pages to cache. */ 294 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 295 296 /* 297 * The btree data structure requires that at least two keys can fit on 298 * a page, but other than that there's no fixed requirement. The user 299 * specified a minimum number per page, and we translated that into the 300 * number of bytes a key/data pair can use before being placed on an 301 * overflow page. This calculation includes the page header, the size 302 * of the index referencing the leaf item and the size of the leaf item 303 * structure. Also, don't let the user specify a minkeypage such that 304 * a key/data pair won't fit even if both key and data are on overflow 305 * pages. 306 */ 307 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 308 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 309 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 310 t->bt_ovflsize = 311 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 312 313 /* Initialize the buffer pool. */ 314 if ((t->bt_mp = 315 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 316 goto err; 317 if (!F_ISSET(t, B_INMEM)) 318 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 319 320 /* Create a root page if new tree. */ 321 if (nroot(t) == RET_ERROR) 322 goto err; 323 324 /* Global flags. */ 325 if (dflags & DB_LOCK) 326 F_SET(t, B_DB_LOCK); 327 if (dflags & DB_SHMEM) 328 F_SET(t, B_DB_SHMEM); 329 if (dflags & DB_TXN) 330 F_SET(t, B_DB_TXN); 331 332 return (dbp); 333 334 einval: errno = EINVAL; 335 goto err; 336 337 eftype: errno = EFTYPE; 338 goto err; 339 340 err: if (t) { 341 if (t->bt_dbp) 342 free(t->bt_dbp); 343 if (t->bt_fd != -1) 344 (void)close(t->bt_fd); 345 free(t); 346 } 347 return (NULL); 348 } 349 350 /* 351 * NROOT -- Create the root of a new tree. 352 * 353 * Parameters: 354 * t: tree 355 * 356 * Returns: 357 * RET_ERROR, RET_SUCCESS 358 */ 359 static int 360 nroot(t) 361 BTREE *t; 362 { 363 PAGE *meta, *root; 364 pgno_t npg; 365 366 if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { 367 mpool_put(t->bt_mp, meta, 0); 368 return (RET_SUCCESS); 369 } 370 if (errno != EINVAL) /* It's OK to not exist. */ 371 return (RET_ERROR); 372 errno = 0; 373 374 if ((meta = mpool_new(t->bt_mp, &npg)) == NULL) 375 return (RET_ERROR); 376 377 if ((root = mpool_new(t->bt_mp, &npg)) == NULL) 378 return (RET_ERROR); 379 380 if (npg != P_ROOT) 381 return (RET_ERROR); 382 root->pgno = npg; 383 root->prevpg = root->nextpg = P_INVALID; 384 root->lower = BTDATAOFF; 385 root->upper = t->bt_psize; 386 root->flags = P_BLEAF; 387 memset(meta, 0, t->bt_psize); 388 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 389 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 390 return (RET_SUCCESS); 391 } 392 393 static int 394 tmp() 395 { 396 sigset_t set, oset; 397 size_t len; 398 int fd; 399 char *envtmp; 400 char path[PATH_MAX]; 401 402 if (issetugid()) 403 envtmp = NULL; 404 else 405 envtmp = getenv("TMPDIR"); 406 407 len = snprintf(path, 408 sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : _PATH_TMP); 409 if (len >= sizeof(path)) 410 return -1; 411 412 (void)sigfillset(&set); 413 (void)sigprocmask(SIG_BLOCK, &set, &oset); 414 if ((fd = mkstemp(path)) != -1) 415 (void)unlink(path); 416 (void)sigprocmask(SIG_SETMASK, &oset, NULL); 417 return(fd); 418 } 419 420 static int 421 byteorder() 422 { 423 u_int32_t x; 424 u_char *p; 425 426 x = 0x01020304; 427 p = (u_char *)(void *)&x; 428 switch (*p) { 429 case 1: 430 return (BIG_ENDIAN); 431 case 4: 432 return (LITTLE_ENDIAN); 433 default: 434 return (0); 435 } 436 } 437 438 int 439 __bt_fd(dbp) 440 const DB *dbp; 441 { 442 BTREE *t; 443 444 t = dbp->internal; 445 446 /* Toss any page pinned across calls. */ 447 if (t->bt_pinned != NULL) { 448 mpool_put(t->bt_mp, t->bt_pinned, 0); 449 t->bt_pinned = NULL; 450 } 451 452 /* In-memory database can't have a file descriptor. */ 453 if (F_ISSET(t, B_INMEM)) { 454 errno = ENOENT; 455 return (-1); 456 } 457 return (t->bt_fd); 458 } 459