1 /* $NetBSD: bt_open.c,v 1.24 2008/09/11 12:58:00 joerg 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 #if HAVE_NBTOOL_CONFIG_H 36 #include "nbtool_config.h" 37 #endif 38 39 #include <sys/cdefs.h> 40 #ifndef __minix 41 __RCSID("$NetBSD: bt_open.c,v 1.24 2008/09/11 12:58:00 joerg Exp $"); 42 #endif 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 #ifndef __minix 53 #include "namespace.h" 54 #endif 55 #include <sys/stat.h> 56 57 #include <assert.h> 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 #ifndef __minix 67 #include <paths.h> 68 #else 69 #define _PATH_TMP "/tmp/" 70 #endif 71 72 #include <db.h> 73 #include "btree.h" 74 75 #ifdef DEBUG 76 #undef MINPSIZE 77 #define MINPSIZE 128 78 #endif 79 80 #ifndef LITTLE_ENDIAN 81 # define LITTLE_ENDIAN 1234 82 #endif 83 84 #ifndef BIG_ENDIAN 85 # define BIG_ENDIAN 4321 86 #endif 87 88 #ifndef BYTE_ORDER 89 #define BYTE_ORDER LITTLE_ENDIAN 90 #endif 91 92 static int byteorder(void); 93 static int nroot(BTREE *); 94 static int tmp(void); 95 96 /* 97 * __BT_OPEN -- Open a btree. 98 * 99 * Creates and fills a DB struct, and calls the routine that actually 100 * opens the btree. 101 * 102 * Parameters: 103 * fname: filename (NULL for in-memory trees) 104 * flags: open flag bits 105 * mode: open permission bits 106 * b: BTREEINFO pointer 107 * 108 * Returns: 109 * NULL on failure, pointer to DB on success. 110 * 111 */ 112 DB * 113 __bt_open(const char *fname, int flags, mode_t mode, const BTREEINFO *openinfo, 114 int dflags) 115 { 116 struct stat sb; 117 BTMETA m; 118 BTREE *t; 119 BTREEINFO b; 120 DB *dbp; 121 pgno_t ncache; 122 ssize_t nr; 123 size_t temp; 124 int machine_lorder; 125 126 t = NULL; 127 128 /* 129 * Intention is to make sure all of the user's selections are okay 130 * here and then use them without checking. Can't be complete, since 131 * we don't know the right page size, lorder or flags until the backing 132 * file is opened. Also, the file's page size can cause the cachesize 133 * to change. 134 */ 135 machine_lorder = byteorder(); 136 if (openinfo) { 137 b = *openinfo; 138 139 /* Flags: R_DUP. */ 140 if (b.flags & ~(R_DUP)) 141 goto einval; 142 143 /* 144 * Page size must be indx_t aligned and >= MINPSIZE. Default 145 * page size is set farther on, based on the underlying file 146 * transfer size. 147 */ 148 if (b.psize && 149 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 150 b.psize & (sizeof(indx_t) - 1))) 151 goto einval; 152 153 /* Minimum number of keys per page; absolute minimum is 2. */ 154 if (b.minkeypage) { 155 if (b.minkeypage < 2) 156 goto einval; 157 } else 158 b.minkeypage = DEFMINKEYPAGE; 159 160 /* If no comparison, use default comparison and prefix. */ 161 if (b.compare == NULL) { 162 b.compare = __bt_defcmp; 163 if (b.prefix == NULL) 164 b.prefix = __bt_defpfx; 165 } 166 167 if (b.lorder == 0) 168 b.lorder = machine_lorder; 169 } else { 170 b.compare = __bt_defcmp; 171 b.cachesize = 0; 172 b.flags = 0; 173 b.lorder = machine_lorder; 174 b.minkeypage = DEFMINKEYPAGE; 175 b.prefix = __bt_defpfx; 176 b.psize = 0; 177 } 178 179 /* Check for the ubiquitous PDP-11. */ 180 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 181 goto einval; 182 183 /* Allocate and initialize DB and BTREE structures. */ 184 if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL) 185 goto err; 186 memset(t, 0, sizeof(BTREE)); 187 t->bt_fd = -1; /* Don't close unopened fd on error. */ 188 t->bt_lorder = b.lorder; 189 t->bt_order = NOT; 190 t->bt_cmp = b.compare; 191 t->bt_pfx = b.prefix; 192 t->bt_rfd = -1; 193 194 if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL) 195 goto err; 196 memset(t->bt_dbp, 0, sizeof(DB)); 197 if (t->bt_lorder != machine_lorder) 198 F_SET(t, B_NEEDSWAP); 199 200 dbp->type = DB_BTREE; 201 dbp->internal = t; 202 dbp->close = __bt_close; 203 dbp->del = __bt_delete; 204 dbp->fd = __bt_fd; 205 dbp->get = __bt_get; 206 dbp->put = __bt_put; 207 dbp->seq = __bt_seq; 208 dbp->sync = __bt_sync; 209 210 /* 211 * If no file name was supplied, this is an in-memory btree and we 212 * open a backing temporary file. Otherwise, it's a disk-based tree. 213 */ 214 if (fname) { 215 switch (flags & O_ACCMODE) { 216 case O_RDONLY: 217 F_SET(t, B_RDONLY); 218 break; 219 case O_RDWR: 220 break; 221 case O_WRONLY: 222 default: 223 goto einval; 224 } 225 226 if ((t->bt_fd = open(fname, flags, mode)) == -1) 227 goto err; 228 if (fcntl(t->bt_fd, F_SETFD, FD_CLOEXEC) == -1) 229 goto err; 230 } else { 231 if ((flags & O_ACCMODE) != O_RDWR) 232 goto einval; 233 if ((t->bt_fd = tmp()) == -1) 234 goto err; 235 F_SET(t, B_INMEM); 236 } 237 238 if (fcntl(t->bt_fd, F_SETFD, FD_CLOEXEC) == -1) 239 goto err; 240 241 if (fstat(t->bt_fd, &sb)) 242 goto err; 243 if (sb.st_size) { 244 if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 245 goto err; 246 if (nr != sizeof(BTMETA)) 247 goto eftype; 248 249 /* 250 * Read in the meta-data. This can change the notion of what 251 * the lorder, page size and flags are, and, when the page size 252 * changes, the cachesize value can change too. If the user 253 * specified the wrong byte order for an existing database, we 254 * don't bother to return an error, we just clear the NEEDSWAP 255 * bit. 256 */ 257 if (m.magic == BTREEMAGIC) 258 F_CLR(t, B_NEEDSWAP); 259 else { 260 F_SET(t, B_NEEDSWAP); 261 M_32_SWAP(m.magic); 262 M_32_SWAP(m.version); 263 M_32_SWAP(m.psize); 264 M_32_SWAP(m.free); 265 M_32_SWAP(m.nrecs); 266 M_32_SWAP(m.flags); 267 } 268 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 269 goto eftype; 270 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 271 m.psize & (sizeof(indx_t) - 1)) 272 goto eftype; 273 if (m.flags & ~SAVEMETA) 274 goto eftype; 275 b.psize = m.psize; 276 F_SET(t, m.flags); 277 t->bt_free = m.free; 278 t->bt_nrecs = m.nrecs; 279 } else { 280 /* 281 * Set the page size to the best value for I/O to this file. 282 * Don't overflow the page offset type. 283 */ 284 if (b.psize == 0) { 285 #ifndef __minix 286 b.psize = sb.st_blksize; 287 #else 288 b.psize = 4096; 289 #endif 290 if (b.psize < MINPSIZE) 291 b.psize = MINPSIZE; 292 if (b.psize > MAX_PAGE_OFFSET + 1) 293 b.psize = MAX_PAGE_OFFSET + 1; 294 } 295 296 /* Set flag if duplicates permitted. */ 297 if (!(b.flags & R_DUP)) 298 F_SET(t, B_NODUPS); 299 300 t->bt_free = P_INVALID; 301 t->bt_nrecs = 0; 302 F_SET(t, B_METADIRTY); 303 } 304 305 t->bt_psize = b.psize; 306 307 /* Set the cache size; must be a multiple of the page size. */ 308 if (b.cachesize && b.cachesize & (b.psize - 1)) 309 b.cachesize += (~b.cachesize & (b.psize - 1)) + 1; 310 if (b.cachesize < b.psize * MINCACHE) 311 b.cachesize = b.psize * MINCACHE; 312 313 /* Calculate number of pages to cache. */ 314 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 315 316 /* 317 * The btree data structure requires that at least two keys can fit on 318 * a page, but other than that there's no fixed requirement. The user 319 * specified a minimum number per page, and we translated that into the 320 * number of bytes a key/data pair can use before being placed on an 321 * overflow page. This calculation includes the page header, the size 322 * of the index referencing the leaf item and the size of the leaf item 323 * structure. Also, don't let the user specify a minkeypage such that 324 * a key/data pair won't fit even if both key and data are on overflow 325 * pages. 326 */ 327 temp = (t->bt_psize - BTDATAOFF) / b.minkeypage - 328 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 329 _DBFIT(temp, indx_t); 330 t->bt_ovflsize = (indx_t)temp; 331 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 332 t->bt_ovflsize = 333 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 334 335 /* Initialize the buffer pool. */ 336 if ((t->bt_mp = 337 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 338 goto err; 339 if (!F_ISSET(t, B_INMEM)) 340 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 341 342 /* Create a root page if new tree. */ 343 if (nroot(t) == RET_ERROR) 344 goto err; 345 346 /* Global flags. */ 347 if (dflags & DB_LOCK) 348 F_SET(t, B_DB_LOCK); 349 if (dflags & DB_SHMEM) 350 F_SET(t, B_DB_SHMEM); 351 if (dflags & DB_TXN) 352 F_SET(t, B_DB_TXN); 353 354 return (dbp); 355 356 einval: errno = EINVAL; 357 goto err; 358 359 eftype: errno = EFTYPE; 360 goto err; 361 362 err: if (t) { 363 if (t->bt_dbp) 364 free(t->bt_dbp); 365 if (t->bt_fd != -1) 366 (void)close(t->bt_fd); 367 free(t); 368 } 369 return (NULL); 370 } 371 372 /* 373 * NROOT -- Create the root of a new tree. 374 * 375 * Parameters: 376 * t: tree 377 * 378 * Returns: 379 * RET_ERROR, RET_SUCCESS 380 */ 381 static int 382 nroot(BTREE *t) 383 { 384 PAGE *meta, *root; 385 pgno_t npg; 386 387 if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { 388 mpool_put(t->bt_mp, meta, 0); 389 return (RET_SUCCESS); 390 } 391 if (errno != EINVAL) /* It's OK to not exist. */ 392 return (RET_ERROR); 393 errno = 0; 394 395 if ((meta = mpool_new(t->bt_mp, &npg)) == NULL) 396 return (RET_ERROR); 397 398 if ((root = mpool_new(t->bt_mp, &npg)) == NULL) 399 return (RET_ERROR); 400 401 if (npg != P_ROOT) 402 return (RET_ERROR); 403 root->pgno = npg; 404 root->prevpg = root->nextpg = P_INVALID; 405 root->lower = BTDATAOFF; 406 root->upper = t->bt_psize; 407 root->flags = P_BLEAF; 408 memset(meta, 0, t->bt_psize); 409 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 410 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 411 return (RET_SUCCESS); 412 } 413 414 static int 415 tmp(void) 416 { 417 sigset_t set, oset; 418 size_t len; 419 int fd; 420 char *envtmp; 421 char path[PATH_MAX]; 422 423 #ifndef __minix 424 if (issetugid()) 425 envtmp = NULL; 426 else 427 envtmp = getenv("TMPDIR"); 428 #else 429 envtmp = getenv("TMPDIR"); 430 #endif 431 432 len = snprintf(path, 433 sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : _PATH_TMP); 434 if (len >= sizeof(path)) 435 return -1; 436 437 (void)sigfillset(&set); 438 (void)sigprocmask(SIG_BLOCK, &set, &oset); 439 if ((fd = mkstemp(path)) != -1) { 440 (void)unlink(path); 441 (void)fcntl(fd, F_SETFD, FD_CLOEXEC); 442 } 443 (void)sigprocmask(SIG_SETMASK, &oset, NULL); 444 return(fd); 445 } 446 447 static int 448 byteorder(void) 449 { 450 uint32_t x; 451 uint8_t *p; 452 453 x = 0x01020304; 454 p = (uint8_t *)(void *)&x; 455 switch (*p) { 456 case 1: 457 return (BIG_ENDIAN); 458 case 4: 459 return (LITTLE_ENDIAN); 460 default: 461 return (0); 462 } 463 } 464 465 int 466 __bt_fd(const DB *dbp) 467 { 468 BTREE *t; 469 470 t = dbp->internal; 471 472 /* Toss any page pinned across calls. */ 473 if (t->bt_pinned != NULL) { 474 mpool_put(t->bt_mp, t->bt_pinned, 0); 475 t->bt_pinned = NULL; 476 } 477 478 /* In-memory database can't have a file descriptor. */ 479 if (F_ISSET(t, B_INMEM)) { 480 errno = ENOENT; 481 return (-1); 482 } 483 return (t->bt_fd); 484 } 485