1 /*- 2 * Copyright (c) 1990, 1993, 1994 3 * The Regents of the University of California. All rights reserved. 4 * 5 * This code is derived from software contributed to Berkeley by 6 * Mike Olson. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 3. All advertising materials mentioning features or use of this software 17 * must display the following acknowledgement: 18 * This product includes software developed by the University of 19 * California, Berkeley and its contributors. 20 * 4. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * $FreeBSD: src/lib/libc/db/btree/bt_open.c,v 1.7.2.1 2000/11/02 10:30:07 kris Exp $ 37 * $DragonFly: src/lib/libc/db/btree/bt_open.c,v 1.4 2005/01/31 22:29:07 dillon Exp $ 38 * 39 * @(#)bt_open.c 8.10 (Berkeley) 8/17/94 40 */ 41 42 /* 43 * Implementation of btree access method for 4.4BSD. 44 * 45 * The design here was originally based on that of the btree access method 46 * used in the Postgres database system at UC Berkeley. This implementation 47 * is wholly independent of the Postgres code. 48 */ 49 50 #include "namespace.h" 51 #include <sys/param.h> 52 #include <sys/stat.h> 53 54 #include <errno.h> 55 #include <fcntl.h> 56 #include <limits.h> 57 #include <signal.h> 58 #include <stdio.h> 59 #include <stdlib.h> 60 #include <string.h> 61 #include <unistd.h> 62 #include "un-namespace.h" 63 64 #include <db.h> 65 #include "btree.h" 66 67 #ifdef DEBUG 68 #undef MINPSIZE 69 #define MINPSIZE 128 70 #endif 71 72 static int byteorder (void); 73 static int nroot (BTREE *); 74 static int tmp (void); 75 76 /* 77 * __BT_OPEN -- Open a btree. 78 * 79 * Creates and fills a DB struct, and calls the routine that actually 80 * opens the btree. 81 * 82 * Parameters: 83 * fname: filename (NULL for in-memory trees) 84 * flags: open flag bits 85 * mode: open permission bits 86 * b: BTREEINFO pointer 87 * 88 * Returns: 89 * NULL on failure, pointer to DB on success. 90 * 91 */ 92 DB * 93 __bt_open(fname, flags, mode, openinfo, dflags) 94 const char *fname; 95 int flags, mode, dflags; 96 const BTREEINFO *openinfo; 97 { 98 struct stat sb; 99 BTMETA m; 100 BTREE *t; 101 BTREEINFO b; 102 DB *dbp; 103 pgno_t ncache; 104 ssize_t nr; 105 int machine_lorder; 106 107 t = NULL; 108 109 /* 110 * Intention is to make sure all of the user's selections are okay 111 * here and then use them without checking. Can't be complete, since 112 * we don't know the right page size, lorder or flags until the backing 113 * file is opened. Also, the file's page size can cause the cachesize 114 * to change. 115 */ 116 machine_lorder = byteorder(); 117 if (openinfo) { 118 b = *openinfo; 119 120 /* Flags: R_DUP. */ 121 if (b.flags & ~(R_DUP)) 122 goto einval; 123 124 /* 125 * Page size must be indx_t aligned and >= MINPSIZE. Default 126 * page size is set farther on, based on the underlying file 127 * transfer size. 128 */ 129 if (b.psize && 130 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 131 b.psize & (sizeof(indx_t) - 1) )) 132 goto einval; 133 134 /* Minimum number of keys per page; absolute minimum is 2. */ 135 if (b.minkeypage) { 136 if (b.minkeypage < 2) 137 goto einval; 138 } else 139 b.minkeypage = DEFMINKEYPAGE; 140 141 /* If no comparison, use default comparison and prefix. */ 142 if (b.compare == NULL) { 143 b.compare = __bt_defcmp; 144 if (b.prefix == NULL) 145 b.prefix = __bt_defpfx; 146 } 147 148 if (b.lorder == 0) 149 b.lorder = machine_lorder; 150 } else { 151 b.compare = __bt_defcmp; 152 b.cachesize = 0; 153 b.flags = 0; 154 b.lorder = machine_lorder; 155 b.minkeypage = DEFMINKEYPAGE; 156 b.prefix = __bt_defpfx; 157 b.psize = 0; 158 } 159 160 /* Check for the ubiquitous PDP-11. */ 161 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 162 goto einval; 163 164 /* Allocate and initialize DB and BTREE structures. */ 165 if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL) 166 goto err; 167 memset(t, 0, sizeof(BTREE)); 168 t->bt_fd = -1; /* Don't close unopened fd on error. */ 169 t->bt_lorder = b.lorder; 170 t->bt_order = NOT; 171 t->bt_cmp = b.compare; 172 t->bt_pfx = b.prefix; 173 t->bt_rfd = -1; 174 175 if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL) 176 goto err; 177 memset(t->bt_dbp, 0, sizeof(DB)); 178 if (t->bt_lorder != machine_lorder) 179 F_SET(t, B_NEEDSWAP); 180 181 dbp->type = DB_BTREE; 182 dbp->internal = t; 183 dbp->close = __bt_close; 184 dbp->del = __bt_delete; 185 dbp->fd = __bt_fd; 186 dbp->get = __bt_get; 187 dbp->put = __bt_put; 188 dbp->seq = __bt_seq; 189 dbp->sync = __bt_sync; 190 191 /* 192 * If no file name was supplied, this is an in-memory btree and we 193 * open a backing temporary file. Otherwise, it's a disk-based tree. 194 */ 195 if (fname) { 196 switch (flags & O_ACCMODE) { 197 case O_RDONLY: 198 F_SET(t, B_RDONLY); 199 break; 200 case O_RDWR: 201 break; 202 case O_WRONLY: 203 default: 204 goto einval; 205 } 206 207 if ((t->bt_fd = _open(fname, flags, mode)) < 0) 208 goto err; 209 210 } else { 211 if ((flags & O_ACCMODE) != O_RDWR) 212 goto einval; 213 if ((t->bt_fd = tmp()) == -1) 214 goto err; 215 F_SET(t, B_INMEM); 216 } 217 218 if (_fcntl(t->bt_fd, F_SETFD, 1) == -1) 219 goto err; 220 221 if (_fstat(t->bt_fd, &sb)) 222 goto err; 223 if (sb.st_size) { 224 if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 225 goto err; 226 if (nr != sizeof(BTMETA)) 227 goto eftype; 228 229 /* 230 * Read in the meta-data. This can change the notion of what 231 * the lorder, page size and flags are, and, when the page size 232 * changes, the cachesize value can change too. If the user 233 * specified the wrong byte order for an existing database, we 234 * don't bother to return an error, we just clear the NEEDSWAP 235 * bit. 236 */ 237 if (m.magic == BTREEMAGIC) 238 F_CLR(t, B_NEEDSWAP); 239 else { 240 F_SET(t, B_NEEDSWAP); 241 M_32_SWAP(m.magic); 242 M_32_SWAP(m.version); 243 M_32_SWAP(m.psize); 244 M_32_SWAP(m.free); 245 M_32_SWAP(m.nrecs); 246 M_32_SWAP(m.flags); 247 } 248 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 249 goto eftype; 250 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 251 m.psize & (sizeof(indx_t) - 1) ) 252 goto eftype; 253 if (m.flags & ~SAVEMETA) 254 goto eftype; 255 b.psize = m.psize; 256 F_SET(t, m.flags); 257 t->bt_free = m.free; 258 t->bt_nrecs = m.nrecs; 259 } else { 260 /* 261 * Set the page size to the best value for I/O to this file. 262 * Don't overflow the page offset type. 263 */ 264 if (b.psize == 0) { 265 b.psize = sb.st_blksize; 266 if (b.psize < MINPSIZE) 267 b.psize = MINPSIZE; 268 if (b.psize > MAX_PAGE_OFFSET + 1) 269 b.psize = MAX_PAGE_OFFSET + 1; 270 } 271 272 /* Set flag if duplicates permitted. */ 273 if (!(b.flags & R_DUP)) 274 F_SET(t, B_NODUPS); 275 276 t->bt_free = P_INVALID; 277 t->bt_nrecs = 0; 278 F_SET(t, B_METADIRTY); 279 } 280 281 t->bt_psize = b.psize; 282 283 /* Set the cache size; must be a multiple of the page size. */ 284 if (b.cachesize && b.cachesize & (b.psize - 1) ) 285 b.cachesize += (~b.cachesize & (b.psize - 1) ) + 1; 286 if (b.cachesize < b.psize * MINCACHE) 287 b.cachesize = b.psize * MINCACHE; 288 289 /* Calculate number of pages to cache. */ 290 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 291 292 /* 293 * The btree data structure requires that at least two keys can fit on 294 * a page, but other than that there's no fixed requirement. The user 295 * specified a minimum number per page, and we translated that into the 296 * number of bytes a key/data pair can use before being placed on an 297 * overflow page. This calculation includes the page header, the size 298 * of the index referencing the leaf item and the size of the leaf item 299 * structure. Also, don't let the user specify a minkeypage such that 300 * a key/data pair won't fit even if both key and data are on overflow 301 * pages. 302 */ 303 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 304 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 305 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 306 t->bt_ovflsize = 307 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 308 309 /* Initialize the buffer pool. */ 310 if ((t->bt_mp = 311 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 312 goto err; 313 if (!F_ISSET(t, B_INMEM)) 314 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 315 316 /* Create a root page if new tree. */ 317 if (nroot(t) == RET_ERROR) 318 goto err; 319 320 /* Global flags. */ 321 if (dflags & DB_LOCK) 322 F_SET(t, B_DB_LOCK); 323 if (dflags & DB_SHMEM) 324 F_SET(t, B_DB_SHMEM); 325 if (dflags & DB_TXN) 326 F_SET(t, B_DB_TXN); 327 328 return (dbp); 329 330 einval: errno = EINVAL; 331 goto err; 332 333 eftype: errno = EFTYPE; 334 goto err; 335 336 err: if (t) { 337 if (t->bt_dbp) 338 free(t->bt_dbp); 339 if (t->bt_fd != -1) 340 (void)_close(t->bt_fd); 341 free(t); 342 } 343 return (NULL); 344 } 345 346 /* 347 * NROOT -- Create the root of a new tree. 348 * 349 * Parameters: 350 * t: tree 351 * 352 * Returns: 353 * RET_ERROR, RET_SUCCESS 354 */ 355 static int 356 nroot(t) 357 BTREE *t; 358 { 359 PAGE *meta, *root; 360 pgno_t npg; 361 362 if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { 363 mpool_put(t->bt_mp, meta, 0); 364 return (RET_SUCCESS); 365 } 366 if (errno != EINVAL) /* It's OK to not exist. */ 367 return (RET_ERROR); 368 errno = 0; 369 370 if ((meta = mpool_new(t->bt_mp, &npg)) == NULL) 371 return (RET_ERROR); 372 373 if ((root = mpool_new(t->bt_mp, &npg)) == NULL) 374 return (RET_ERROR); 375 376 if (npg != P_ROOT) 377 return (RET_ERROR); 378 root->pgno = npg; 379 root->prevpg = root->nextpg = P_INVALID; 380 root->lower = BTDATAOFF; 381 root->upper = t->bt_psize; 382 root->flags = P_BLEAF; 383 memset(meta, 0, t->bt_psize); 384 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 385 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 386 return (RET_SUCCESS); 387 } 388 389 static int 390 tmp() 391 { 392 sigset_t set, oset; 393 int fd; 394 char *envtmp = NULL; 395 char path[MAXPATHLEN]; 396 397 if (issetugid() == 0) 398 envtmp = getenv("TMPDIR"); 399 (void)snprintf(path, 400 sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp"); 401 402 (void)sigfillset(&set); 403 (void)_sigprocmask(SIG_BLOCK, &set, &oset); 404 if ((fd = mkstemp(path)) != -1) 405 (void)unlink(path); 406 (void)_sigprocmask(SIG_SETMASK, &oset, NULL); 407 return(fd); 408 } 409 410 static int 411 byteorder() 412 { 413 u_int32_t x; 414 u_char *p; 415 416 x = 0x01020304; 417 p = (u_char *)&x; 418 switch (*p) { 419 case 1: 420 return (BIG_ENDIAN); 421 case 4: 422 return (LITTLE_ENDIAN); 423 default: 424 return (0); 425 } 426 } 427 428 int 429 __bt_fd(dbp) 430 const DB *dbp; 431 { 432 BTREE *t; 433 434 t = dbp->internal; 435 436 /* Toss any page pinned across calls. */ 437 if (t->bt_pinned != NULL) { 438 mpool_put(t->bt_mp, t->bt_pinned, 0); 439 t->bt_pinned = NULL; 440 } 441 442 /* In-memory database can't have a file descriptor. */ 443 if (F_ISSET(t, B_INMEM)) { 444 errno = ENOENT; 445 return (-1); 446 } 447 return (t->bt_fd); 448 } 449