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 static char sccsid[] = "@(#)bt_open.c 8.1 (Berkeley) 6/4/93"; 39 #endif /* LIBC_SCCS and not lint */ 40 41 /* 42 * Implementation of btree access method for 4.4BSD. 43 * 44 * The design here was originally based on that of the btree access method 45 * used in the Postgres database system at UC Berkeley. This implementation 46 * is wholly independent of the Postgres code. 47 */ 48 49 #include <sys/param.h> 50 #include <sys/stat.h> 51 52 #include <errno.h> 53 #include <fcntl.h> 54 #include <limits.h> 55 #include <signal.h> 56 #include <stdio.h> 57 #include <stdlib.h> 58 #include <string.h> 59 #include <unistd.h> 60 61 #define __DBINTERFACE_PRIVATE 62 #include <db.h> 63 #include "btree.h" 64 65 static int byteorder __P((void)); 66 static int nroot __P((BTREE *)); 67 static int tmp __P((void)); 68 69 /* 70 * __BT_OPEN -- Open a btree. 71 * 72 * Creates and fills a DB struct, and calls the routine that actually 73 * opens the btree. 74 * 75 * Parameters: 76 * fname: filename (NULL for in-memory trees) 77 * flags: open flag bits 78 * mode: open permission bits 79 * b: BTREEINFO pointer 80 * 81 * Returns: 82 * NULL on failure, pointer to DB on success. 83 * 84 */ 85 DB * 86 __bt_open(fname, flags, mode, openinfo) 87 const char *fname; 88 int flags, mode; 89 const BTREEINFO *openinfo; 90 { 91 BTMETA m; 92 BTREE *t; 93 BTREEINFO b; 94 DB *dbp; 95 pgno_t ncache; 96 struct stat sb; 97 int machine_lorder, nr; 98 99 t = NULL; 100 101 /* 102 * Intention is to make sure all of the user's selections are okay 103 * here and then use them without checking. Can't be complete, since 104 * we don't know the right page size, lorder or flags until the backing 105 * file is opened. Also, the file's page size can cause the cachesize 106 * to change. 107 */ 108 machine_lorder = byteorder(); 109 if (openinfo) { 110 b = *openinfo; 111 112 /* Flags: R_DUP. */ 113 if (b.flags & ~(R_DUP)) 114 goto einval; 115 116 /* 117 * Page size must be indx_t aligned and >= MINPSIZE. Default 118 * page size is set farther on, based on the underlying file 119 * transfer size. 120 */ 121 if (b.psize && 122 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 123 b.psize & sizeof(indx_t) - 1)) 124 goto einval; 125 126 /* Minimum number of keys per page; absolute minimum is 2. */ 127 if (b.minkeypage) { 128 if (b.minkeypage < 2) 129 goto einval; 130 } else 131 b.minkeypage = DEFMINKEYPAGE; 132 133 /* If no comparison, use default comparison and prefix. */ 134 if (b.compare == NULL) { 135 b.compare = __bt_defcmp; 136 if (b.prefix == NULL) 137 b.prefix = __bt_defpfx; 138 } 139 140 if (b.lorder == 0) 141 b.lorder = machine_lorder; 142 } else { 143 b.compare = __bt_defcmp; 144 b.cachesize = 0; 145 b.flags = 0; 146 b.lorder = machine_lorder; 147 b.minkeypage = DEFMINKEYPAGE; 148 b.prefix = __bt_defpfx; 149 b.psize = 0; 150 } 151 152 /* Check for the ubiquitous PDP-11. */ 153 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 154 goto einval; 155 156 /* Allocate and initialize DB and BTREE structures. */ 157 if ((t = malloc(sizeof(BTREE))) == NULL) 158 goto err; 159 t->bt_fd = -1; /* Don't close unopened fd on error. */ 160 if ((t->bt_dbp = dbp = malloc(sizeof(DB))) == NULL) 161 goto err; 162 t->bt_bcursor.pgno = P_INVALID; 163 t->bt_bcursor.index = 0; 164 t->bt_stack = NULL; 165 t->bt_sp = t->bt_maxstack = 0; 166 t->bt_kbuf = t->bt_dbuf = NULL; 167 t->bt_kbufsz = t->bt_dbufsz = 0; 168 t->bt_lorder = b.lorder; 169 t->bt_order = NOT; 170 t->bt_cmp = b.compare; 171 t->bt_pfx = b.prefix; 172 t->bt_flags = 0; 173 if (t->bt_lorder != machine_lorder) 174 SET(t, B_NEEDSWAP); 175 176 dbp->type = DB_BTREE; 177 dbp->internal = t; 178 dbp->close = __bt_close; 179 dbp->del = __bt_delete; 180 dbp->fd = __bt_fd; 181 dbp->get = __bt_get; 182 dbp->put = __bt_put; 183 dbp->seq = __bt_seq; 184 dbp->sync = __bt_sync; 185 186 /* 187 * If no file name was supplied, this is an in-memory btree and we 188 * open a backing temporary file. Otherwise, it's a disk-based tree. 189 */ 190 if (fname) { 191 switch(flags & O_ACCMODE) { 192 case O_RDONLY: 193 SET(t, B_RDONLY); 194 break; 195 case O_RDWR: 196 break; 197 case O_WRONLY: 198 default: 199 goto einval; 200 } 201 202 if ((t->bt_fd = 203 open(fname, flags & __USE_OPEN_FLAGS, mode)) < 0) 204 goto err; 205 206 } else { 207 if ((flags & O_ACCMODE) != O_RDWR) 208 goto einval; 209 if ((t->bt_fd = tmp()) == -1) 210 goto err; 211 SET(t, B_INMEM); 212 } 213 214 if (fcntl(t->bt_fd, F_SETFD, 1) == -1) 215 goto err; 216 217 if (fstat(t->bt_fd, &sb)) 218 goto err; 219 if (sb.st_size) { 220 nr = read(t->bt_fd, &m, sizeof(BTMETA)); 221 if (nr < 0) 222 goto err; 223 if (nr != sizeof(BTMETA)) 224 goto eftype; 225 226 /* 227 * Read in the meta-data. This can change the notion of what 228 * the lorder, page size and flags are, and, when the page size 229 * changes, the cachesize value can change too. If the user 230 * specified the wrong byte order for an existing database, we 231 * don't bother to return an error, we just clear the NEEDSWAP 232 * bit. 233 */ 234 if (m.m_magic == BTREEMAGIC) 235 CLR(t, B_NEEDSWAP); 236 else { 237 SET(t, B_NEEDSWAP); 238 BLSWAP(m.m_magic); 239 BLSWAP(m.m_version); 240 BLSWAP(m.m_psize); 241 BLSWAP(m.m_free); 242 BLSWAP(m.m_nrecs); 243 BLSWAP(m.m_flags); 244 } 245 if (m.m_magic != BTREEMAGIC || m.m_version != BTREEVERSION) 246 goto eftype; 247 if (m.m_psize < MINPSIZE || m.m_psize > MAX_PAGE_OFFSET + 1 || 248 m.m_psize & sizeof(indx_t) - 1) 249 goto eftype; 250 if (m.m_flags & ~SAVEMETA) 251 goto eftype; 252 b.psize = m.m_psize; 253 t->bt_flags |= m.m_flags; 254 t->bt_free = m.m_free; 255 t->bt_nrecs = m.m_nrecs; 256 } else { 257 /* 258 * Set the page size to the best value for I/O to this file. 259 * Don't overflow the page offset type. 260 */ 261 if (b.psize == 0) { 262 b.psize = sb.st_blksize; 263 if (b.psize < MINPSIZE) 264 b.psize = MINPSIZE; 265 if (b.psize > MAX_PAGE_OFFSET + 1) 266 b.psize = MAX_PAGE_OFFSET + 1; 267 } 268 269 /* Set flag if duplicates permitted. */ 270 if (!(b.flags & R_DUP)) 271 SET(t, B_NODUPS); 272 273 t->bt_free = P_INVALID; 274 t->bt_nrecs = 0; 275 SET(t, B_METADIRTY); 276 } 277 278 t->bt_psize = b.psize; 279 280 /* Set the cache size; must be a multiple of the page size. */ 281 if (b.cachesize && b.cachesize & b.psize - 1) 282 b.cachesize += (~b.cachesize & b.psize - 1) + 1; 283 if (b.cachesize < b.psize * MINCACHE) 284 b.cachesize = b.psize * MINCACHE; 285 286 /* Calculate number of pages to cache. */ 287 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 288 289 /* 290 * The btree data structure requires that at least two keys can fit on 291 * a page, but other than that there's no fixed requirement. The user 292 * specified a minimum number per page, and we translated that into the 293 * number of bytes a key/data pair can use before being placed on an 294 * overflow page. This calculation includes the page header, the size 295 * of the index referencing the leaf item and the size of the leaf item 296 * structure. Also, don't let the user specify a minkeypage such that 297 * a key/data pair won't fit even if both key and data are on overflow 298 * pages. 299 */ 300 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 301 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 302 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 303 t->bt_ovflsize = 304 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 305 306 /* Initialize the buffer pool. */ 307 if ((t->bt_mp = 308 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 309 goto err; 310 if (!ISSET(t, B_INMEM)) 311 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 312 313 /* Create a root page if new tree. */ 314 if (nroot(t) == RET_ERROR) 315 goto err; 316 317 return (dbp); 318 319 einval: errno = EINVAL; 320 goto err; 321 322 eftype: errno = EFTYPE; 323 goto err; 324 325 err: if (t) { 326 if (t->bt_dbp) 327 free(t->bt_dbp); 328 if (t->bt_fd != -1) 329 (void)close(t->bt_fd); 330 free(t); 331 } 332 return (NULL); 333 } 334 335 /* 336 * NROOT -- Create the root of a new tree. 337 * 338 * Parameters: 339 * t: tree 340 * 341 * Returns: 342 * RET_ERROR, RET_SUCCESS 343 */ 344 static int 345 nroot(t) 346 BTREE *t; 347 { 348 PAGE *meta, *root; 349 pgno_t npg; 350 351 if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { 352 mpool_put(t->bt_mp, meta, 0); 353 return (RET_SUCCESS); 354 } 355 if (errno != EINVAL) 356 return (RET_ERROR); 357 358 if ((meta = mpool_new(t->bt_mp, &npg)) == NULL) 359 return (RET_ERROR); 360 361 if ((root = mpool_new(t->bt_mp, &npg)) == NULL) 362 return (RET_ERROR); 363 364 if (npg != P_ROOT) 365 return (RET_ERROR); 366 root->pgno = npg; 367 root->prevpg = root->nextpg = P_INVALID; 368 root->lower = BTDATAOFF; 369 root->upper = t->bt_psize; 370 root->flags = P_BLEAF; 371 memset(meta, 0, t->bt_psize); 372 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 373 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 374 return (RET_SUCCESS); 375 } 376 377 static int 378 tmp() 379 { 380 sigset_t set, oset; 381 int fd; 382 char *envtmp; 383 char path[MAXPATHLEN]; 384 385 envtmp = getenv("TMPDIR"); 386 (void)snprintf(path, 387 sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : "/tmp"); 388 389 (void)sigfillset(&set); 390 (void)sigprocmask(SIG_BLOCK, &set, &oset); 391 if ((fd = mkstemp(path)) != -1) 392 (void)unlink(path); 393 (void)sigprocmask(SIG_SETMASK, &oset, NULL); 394 return(fd); 395 } 396 397 static int 398 byteorder() 399 { 400 u_long x; /* XXX: 32-bit assumption. */ 401 u_char *p; 402 403 x = 0x01020304; 404 p = (u_char *)&x; 405 switch (*p) { 406 case 1: 407 return (BIG_ENDIAN); 408 case 4: 409 return (LITTLE_ENDIAN); 410 default: 411 return (0); 412 } 413 } 414 415 int 416 __bt_fd(dbp) 417 const DB *dbp; 418 { 419 BTREE *t; 420 421 t = dbp->internal; 422 423 if (ISSET(t, B_INMEM)) { 424 errno = ENOENT; 425 return (-1); 426 } 427 return (t->bt_fd); 428 } 429