1 /*- 2 * Copyright (c) 1990, 1993 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 37 #if defined(LIBC_SCCS) && !defined(lint) 38 /* from: static char sccsid[] = "@(#)bt_open.c 8.3 (Berkeley) 9/16/93"; */ 39 static char *rcsid = "$Id: bt_open.c,v 1.5 1993/09/17 01:06:23 cgd Exp $"; 40 #endif /* LIBC_SCCS and not lint */ 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 <sys/param.h> 51 #include <sys/stat.h> 52 53 #include <errno.h> 54 #include <fcntl.h> 55 #include <limits.h> 56 #include <signal.h> 57 #include <stdio.h> 58 #include <stdlib.h> 59 #include <string.h> 60 #include <unistd.h> 61 62 #define __DBINTERFACE_PRIVATE 63 #include <db.h> 64 #include "btree.h" 65 66 static int byteorder __P((void)); 67 static int nroot __P((BTREE *)); 68 static int tmp __P((void)); 69 70 /* 71 * __BT_OPEN -- Open a btree. 72 * 73 * Creates and fills a DB struct, and calls the routine that actually 74 * opens the btree. 75 * 76 * Parameters: 77 * fname: filename (NULL for in-memory trees) 78 * flags: open flag bits 79 * mode: open permission bits 80 * b: BTREEINFO pointer 81 * 82 * Returns: 83 * NULL on failure, pointer to DB on success. 84 * 85 */ 86 DB * 87 __bt_open(fname, flags, mode, openinfo, dflags) 88 const char *fname; 89 int flags, mode, dflags; 90 const BTREEINFO *openinfo; 91 { 92 BTMETA m; 93 BTREE *t; 94 BTREEINFO b; 95 DB *dbp; 96 pgno_t ncache; 97 struct stat sb; 98 int machine_lorder, nr; 99 100 t = NULL; 101 102 /* 103 * Intention is to make sure all of the user's selections are okay 104 * here and then use them without checking. Can't be complete, since 105 * we don't know the right page size, lorder or flags until the backing 106 * file is opened. Also, the file's page size can cause the cachesize 107 * to change. 108 */ 109 machine_lorder = byteorder(); 110 if (openinfo) { 111 b = *openinfo; 112 113 /* Flags: R_DUP. */ 114 if (b.flags & ~(R_DUP)) 115 goto einval; 116 117 /* 118 * Page size must be indx_t aligned and >= MINPSIZE. Default 119 * page size is set farther on, based on the underlying file 120 * transfer size. 121 */ 122 if (b.psize && 123 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 124 b.psize & sizeof(indx_t) - 1)) 125 goto einval; 126 127 /* Minimum number of keys per page; absolute minimum is 2. */ 128 if (b.minkeypage) { 129 if (b.minkeypage < 2) 130 goto einval; 131 } else 132 b.minkeypage = DEFMINKEYPAGE; 133 134 /* If no comparison, use default comparison and prefix. */ 135 if (b.compare == NULL) { 136 b.compare = __bt_defcmp; 137 if (b.prefix == NULL) 138 b.prefix = __bt_defpfx; 139 } 140 141 if (b.lorder == 0) 142 b.lorder = machine_lorder; 143 } else { 144 b.compare = __bt_defcmp; 145 b.cachesize = 0; 146 b.flags = 0; 147 b.lorder = machine_lorder; 148 b.minkeypage = DEFMINKEYPAGE; 149 b.prefix = __bt_defpfx; 150 b.psize = 0; 151 } 152 153 /* Check for the ubiquitous PDP-11. */ 154 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 155 goto einval; 156 157 /* Allocate and initialize DB and BTREE structures. */ 158 if ((t = malloc(sizeof(BTREE))) == NULL) 159 goto err; 160 memset(t, 0, sizeof(BTREE)); 161 t->bt_bcursor.pgno = P_INVALID; 162 t->bt_fd = -1; /* Don't close unopened fd on error. */ 163 t->bt_lorder = b.lorder; 164 t->bt_order = NOT; 165 t->bt_cmp = b.compare; 166 t->bt_pfx = b.prefix; 167 t->bt_rfd = -1; 168 169 if ((t->bt_dbp = dbp = malloc(sizeof(DB))) == NULL) 170 goto err; 171 t->bt_flags = 0; 172 if (t->bt_lorder != machine_lorder) 173 SET(t, B_NEEDSWAP); 174 175 dbp->type = DB_BTREE; 176 dbp->internal = t; 177 dbp->close = __bt_close; 178 dbp->del = __bt_delete; 179 dbp->fd = __bt_fd; 180 dbp->get = __bt_get; 181 dbp->put = __bt_put; 182 dbp->seq = __bt_seq; 183 dbp->sync = __bt_sync; 184 185 /* 186 * If no file name was supplied, this is an in-memory btree and we 187 * open a backing temporary file. Otherwise, it's a disk-based tree. 188 */ 189 if (fname) { 190 switch(flags & O_ACCMODE) { 191 case O_RDONLY: 192 SET(t, B_RDONLY); 193 break; 194 case O_RDWR: 195 break; 196 case O_WRONLY: 197 default: 198 goto einval; 199 } 200 201 if ((t->bt_fd = open(fname, flags, mode)) < 0) 202 goto err; 203 204 } else { 205 if ((flags & O_ACCMODE) != O_RDWR) 206 goto einval; 207 if ((t->bt_fd = tmp()) == -1) 208 goto err; 209 SET(t, B_INMEM); 210 } 211 212 if (fcntl(t->bt_fd, F_SETFD, 1) == -1) 213 goto err; 214 215 if (fstat(t->bt_fd, &sb)) 216 goto err; 217 if (sb.st_size) { 218 nr = read(t->bt_fd, &m, sizeof(BTMETA)); 219 if (nr < 0) 220 goto err; 221 if (nr != sizeof(BTMETA)) 222 goto eftype; 223 224 /* 225 * Read in the meta-data. This can change the notion of what 226 * the lorder, page size and flags are, and, when the page size 227 * changes, the cachesize value can change too. If the user 228 * specified the wrong byte order for an existing database, we 229 * don't bother to return an error, we just clear the NEEDSWAP 230 * bit. 231 */ 232 if (m.m_magic == BTREEMAGIC) 233 CLR(t, B_NEEDSWAP); 234 else { 235 SET(t, B_NEEDSWAP); 236 BLSWAP(m.m_magic); 237 BLSWAP(m.m_version); 238 BLSWAP(m.m_psize); 239 BLSWAP(m.m_free); 240 BLSWAP(m.m_nrecs); 241 BLSWAP(m.m_flags); 242 } 243 if (m.m_magic != BTREEMAGIC || m.m_version != BTREEVERSION) 244 goto eftype; 245 if (m.m_psize < MINPSIZE || m.m_psize > MAX_PAGE_OFFSET + 1 || 246 m.m_psize & sizeof(indx_t) - 1) 247 goto eftype; 248 if (m.m_flags & ~SAVEMETA) 249 goto eftype; 250 b.psize = m.m_psize; 251 t->bt_flags |= m.m_flags; 252 t->bt_free = m.m_free; 253 t->bt_nrecs = m.m_nrecs; 254 } else { 255 /* 256 * Set the page size to the best value for I/O to this file. 257 * Don't overflow the page offset type. 258 */ 259 if (b.psize == 0) { 260 b.psize = sb.st_blksize; 261 if (b.psize < MINPSIZE) 262 b.psize = MINPSIZE; 263 if (b.psize > MAX_PAGE_OFFSET + 1) 264 b.psize = MAX_PAGE_OFFSET + 1; 265 } 266 267 /* Set flag if duplicates permitted. */ 268 if (!(b.flags & R_DUP)) 269 SET(t, B_NODUPS); 270 271 t->bt_free = P_INVALID; 272 t->bt_nrecs = 0; 273 SET(t, B_METADIRTY); 274 } 275 276 t->bt_psize = b.psize; 277 278 /* Set the cache size; must be a multiple of the page size. */ 279 if (b.cachesize && b.cachesize & b.psize - 1) 280 b.cachesize += (~b.cachesize & b.psize - 1) + 1; 281 if (b.cachesize < b.psize * MINCACHE) 282 b.cachesize = b.psize * MINCACHE; 283 284 /* Calculate number of pages to cache. */ 285 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 286 287 /* 288 * The btree data structure requires that at least two keys can fit on 289 * a page, but other than that there's no fixed requirement. The user 290 * specified a minimum number per page, and we translated that into the 291 * number of bytes a key/data pair can use before being placed on an 292 * overflow page. This calculation includes the page header, the size 293 * of the index referencing the leaf item and the size of the leaf item 294 * structure. Also, don't let the user specify a minkeypage such that 295 * a key/data pair won't fit even if both key and data are on overflow 296 * pages. 297 */ 298 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 299 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 300 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 301 t->bt_ovflsize = 302 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 303 304 /* Initialize the buffer pool. */ 305 if ((t->bt_mp = 306 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 307 goto err; 308 if (!ISSET(t, B_INMEM)) 309 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 310 311 /* Create a root page if new tree. */ 312 if (nroot(t) == RET_ERROR) 313 goto err; 314 315 /* Global flags. */ 316 if (dflags & DB_LOCK) 317 SET(t, B_DB_LOCK); 318 if (dflags & DB_SHMEM) 319 SET(t, B_DB_SHMEM); 320 if (dflags & DB_TXN) 321 SET(t, B_DB_TXN); 322 323 return (dbp); 324 325 einval: errno = EINVAL; 326 goto err; 327 328 eftype: errno = EFTYPE; 329 goto err; 330 331 err: if (t) { 332 if (t->bt_dbp) 333 free(t->bt_dbp); 334 if (t->bt_fd != -1) 335 (void)close(t->bt_fd); 336 free(t); 337 } 338 return (NULL); 339 } 340 341 /* 342 * NROOT -- Create the root of a new tree. 343 * 344 * Parameters: 345 * t: tree 346 * 347 * Returns: 348 * RET_ERROR, RET_SUCCESS 349 */ 350 static int 351 nroot(t) 352 BTREE *t; 353 { 354 PAGE *meta, *root; 355 pgno_t npg; 356 357 if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { 358 mpool_put(t->bt_mp, meta, 0); 359 return (RET_SUCCESS); 360 } 361 if (errno != EINVAL) 362 return (RET_ERROR); 363 364 if ((meta = mpool_new(t->bt_mp, &npg)) == NULL) 365 return (RET_ERROR); 366 367 if ((root = mpool_new(t->bt_mp, &npg)) == NULL) 368 return (RET_ERROR); 369 370 if (npg != P_ROOT) 371 return (RET_ERROR); 372 root->pgno = npg; 373 root->prevpg = root->nextpg = P_INVALID; 374 root->lower = BTDATAOFF; 375 root->upper = t->bt_psize; 376 root->flags = P_BLEAF; 377 memset(meta, 0, t->bt_psize); 378 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 379 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 380 return (RET_SUCCESS); 381 } 382 383 static int 384 tmp() 385 { 386 sigset_t set, oset; 387 int fd; 388 char *envtmp; 389 char path[MAXPATHLEN]; 390 391 envtmp = getenv("TMPDIR"); 392 (void)snprintf(path, 393 sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : "/tmp"); 394 395 (void)sigfillset(&set); 396 (void)sigprocmask(SIG_BLOCK, &set, &oset); 397 if ((fd = mkstemp(path)) != -1) 398 (void)unlink(path); 399 (void)sigprocmask(SIG_SETMASK, &oset, NULL); 400 return(fd); 401 } 402 403 static int 404 byteorder() 405 { 406 u_long x; /* XXX: 32-bit assumption. */ 407 u_char *p; 408 409 x = 0x01020304; 410 p = (u_char *)&x; 411 switch (*p) { 412 case 1: 413 return (BIG_ENDIAN); 414 case 4: 415 return (LITTLE_ENDIAN); 416 default: 417 return (0); 418 } 419 } 420 421 int 422 __bt_fd(dbp) 423 const DB *dbp; 424 { 425 BTREE *t; 426 427 t = dbp->internal; 428 429 /* Toss any page pinned across calls. */ 430 if (t->bt_pinned != NULL) { 431 mpool_put(t->bt_mp, t->bt_pinned, 0); 432 t->bt_pinned = NULL; 433 } 434 435 /* In-memory database can't have a file descriptor. */ 436 if (ISSET(t, B_INMEM)) { 437 errno = ENOENT; 438 return (-1); 439 } 440 return (t->bt_fd); 441 } 442