bt_open.c revision 8870
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) 38static char sccsid[] = "@(#)bt_open.c 8.5 (Berkeley) 2/21/94"; 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#include <db.h> 62#include "btree.h" 63 64static int byteorder __P((void)); 65static int nroot __P((BTREE *)); 66static int tmp __P((void)); 67 68/* 69 * __BT_OPEN -- Open a btree. 70 * 71 * Creates and fills a DB struct, and calls the routine that actually 72 * opens the btree. 73 * 74 * Parameters: 75 * fname: filename (NULL for in-memory trees) 76 * flags: open flag bits 77 * mode: open permission bits 78 * b: BTREEINFO pointer 79 * 80 * Returns: 81 * NULL on failure, pointer to DB on success. 82 * 83 */ 84DB * 85__bt_open(fname, flags, mode, openinfo, dflags) 86 const char *fname; 87 int flags, mode, dflags; 88 const BTREEINFO *openinfo; 89{ 90 struct stat sb; 91 BTMETA m; 92 BTREE *t; 93 BTREEINFO b; 94 DB *dbp; 95 pgno_t ncache; 96 ssize_t nr; 97 int machine_lorder; 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 = (BTREE *)malloc(sizeof(BTREE))) == NULL) 158 goto err; 159 memset(t, 0, sizeof(BTREE)); 160 t->bt_bcursor.pgno = P_INVALID; 161 t->bt_fd = -1; /* Don't close unopened fd on error. */ 162 t->bt_lorder = b.lorder; 163 t->bt_order = NOT; 164 t->bt_cmp = b.compare; 165 t->bt_pfx = b.prefix; 166 t->bt_rfd = -1; 167 168 if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL) 169 goto err; 170 t->bt_flags = 0; 171 if (t->bt_lorder != machine_lorder) 172 SET(t, B_NEEDSWAP); 173 174 dbp->type = DB_BTREE; 175 dbp->internal = t; 176 dbp->close = __bt_close; 177 dbp->del = __bt_delete; 178 dbp->fd = __bt_fd; 179 dbp->get = __bt_get; 180 dbp->put = __bt_put; 181 dbp->seq = __bt_seq; 182 dbp->sync = __bt_sync; 183 184 /* 185 * If no file name was supplied, this is an in-memory btree and we 186 * open a backing temporary file. Otherwise, it's a disk-based tree. 187 */ 188 if (fname) { 189 switch(flags & O_ACCMODE) { 190 case O_RDONLY: 191 SET(t, B_RDONLY); 192 break; 193 case O_RDWR: 194 break; 195 case O_WRONLY: 196 default: 197 goto einval; 198 } 199 200 if ((t->bt_fd = open(fname, flags, mode)) < 0) 201 goto err; 202 203 } else { 204 if ((flags & O_ACCMODE) != O_RDWR) 205 goto einval; 206 if ((t->bt_fd = tmp()) == -1) 207 goto err; 208 SET(t, B_INMEM); 209 } 210 211 if (fcntl(t->bt_fd, F_SETFD, 1) == -1) 212 goto err; 213 214 if (fstat(t->bt_fd, &sb)) 215 goto err; 216 if (sb.st_size) { 217 nr = read(t->bt_fd, &m, sizeof(BTMETA)); 218 if (nr < 0) 219 goto err; 220 if (nr != sizeof(BTMETA)) 221 goto eftype; 222 223 /* 224 * Read in the meta-data. This can change the notion of what 225 * the lorder, page size and flags are, and, when the page size 226 * changes, the cachesize value can change too. If the user 227 * specified the wrong byte order for an existing database, we 228 * don't bother to return an error, we just clear the NEEDSWAP 229 * bit. 230 */ 231 if (m.m_magic == BTREEMAGIC) 232 CLR(t, B_NEEDSWAP); 233 else { 234 SET(t, B_NEEDSWAP); 235 M_32_SWAP(m.m_magic); 236 M_32_SWAP(m.m_version); 237 M_32_SWAP(m.m_psize); 238 M_32_SWAP(m.m_free); 239 M_32_SWAP(m.m_nrecs); 240 M_32_SWAP(m.m_flags); 241 } 242 if (m.m_magic != BTREEMAGIC || m.m_version != BTREEVERSION) 243 goto eftype; 244 if (m.m_psize < MINPSIZE || m.m_psize > MAX_PAGE_OFFSET + 1 || 245 m.m_psize & sizeof(indx_t) - 1) 246 goto eftype; 247 if (m.m_flags & ~SAVEMETA) 248 goto eftype; 249 b.psize = m.m_psize; 250 t->bt_flags |= m.m_flags; 251 t->bt_free = m.m_free; 252 t->bt_nrecs = m.m_nrecs; 253 } else { 254 /* 255 * Set the page size to the best value for I/O to this file. 256 * Don't overflow the page offset type. 257 */ 258 if (b.psize == 0) { 259 b.psize = sb.st_blksize; 260 if (b.psize < MINPSIZE) 261 b.psize = MINPSIZE; 262 if (b.psize > MAX_PAGE_OFFSET + 1) 263 b.psize = MAX_PAGE_OFFSET + 1; 264 } 265 266 /* Set flag if duplicates permitted. */ 267 if (!(b.flags & R_DUP)) 268 SET(t, B_NODUPS); 269 270 t->bt_free = P_INVALID; 271 t->bt_nrecs = 0; 272 SET(t, B_METADIRTY); 273 } 274 275 t->bt_psize = b.psize; 276 277 /* Set the cache size; must be a multiple of the page size. */ 278 if (b.cachesize && b.cachesize & b.psize - 1) 279 b.cachesize += (~b.cachesize & b.psize - 1) + 1; 280 if (b.cachesize < b.psize * MINCACHE) 281 b.cachesize = b.psize * MINCACHE; 282 283 /* Calculate number of pages to cache. */ 284 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 285 286 /* 287 * The btree data structure requires that at least two keys can fit on 288 * a page, but other than that there's no fixed requirement. The user 289 * specified a minimum number per page, and we translated that into the 290 * number of bytes a key/data pair can use before being placed on an 291 * overflow page. This calculation includes the page header, the size 292 * of the index referencing the leaf item and the size of the leaf item 293 * structure. Also, don't let the user specify a minkeypage such that 294 * a key/data pair won't fit even if both key and data are on overflow 295 * pages. 296 */ 297 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 298 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 299 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 300 t->bt_ovflsize = 301 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 302 303 /* Initialize the buffer pool. */ 304 if ((t->bt_mp = 305 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 306 goto err; 307 if (!ISSET(t, B_INMEM)) 308 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 309 310 /* Create a root page if new tree. */ 311 if (nroot(t) == RET_ERROR) 312 goto err; 313 314 /* Global flags. */ 315 if (dflags & DB_LOCK) 316 SET(t, B_DB_LOCK); 317 if (dflags & DB_SHMEM) 318 SET(t, B_DB_SHMEM); 319 if (dflags & DB_TXN) 320 SET(t, B_DB_TXN); 321 322 return (dbp); 323 324einval: errno = EINVAL; 325 goto err; 326 327eftype: errno = EFTYPE; 328 goto err; 329 330err: if (t) { 331 if (t->bt_dbp) 332 free(t->bt_dbp); 333 if (t->bt_fd != -1) 334 (void)close(t->bt_fd); 335 free(t); 336 } 337 return (NULL); 338} 339 340/* 341 * NROOT -- Create the root of a new tree. 342 * 343 * Parameters: 344 * t: tree 345 * 346 * Returns: 347 * RET_ERROR, RET_SUCCESS 348 */ 349static int 350nroot(t) 351 BTREE *t; 352{ 353 PAGE *meta, *root; 354 pgno_t npg; 355 356 if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { 357 mpool_put(t->bt_mp, meta, 0); 358 return (RET_SUCCESS); 359 } 360 if (errno != EINVAL) 361 return (RET_ERROR); 362 363 if ((meta = mpool_new(t->bt_mp, &npg)) == NULL) 364 return (RET_ERROR); 365 366 if ((root = mpool_new(t->bt_mp, &npg)) == NULL) 367 return (RET_ERROR); 368 369 if (npg != P_ROOT) 370 return (RET_ERROR); 371 root->pgno = npg; 372 root->prevpg = root->nextpg = P_INVALID; 373 root->lower = BTDATAOFF; 374 root->upper = t->bt_psize; 375 root->flags = P_BLEAF; 376 memset(meta, 0, t->bt_psize); 377 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 378 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 379 return (RET_SUCCESS); 380} 381 382static int 383tmp() 384{ 385 sigset_t set, oset; 386 int fd; 387 char *envtmp; 388 char path[MAXPATHLEN]; 389 390 envtmp = getenv("TMPDIR"); 391 (void)snprintf(path, 392 sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : "/tmp"); 393 394 (void)sigfillset(&set); 395 (void)sigprocmask(SIG_BLOCK, &set, &oset); 396 if ((fd = mkstemp(path)) != -1) 397 (void)unlink(path); 398 (void)sigprocmask(SIG_SETMASK, &oset, NULL); 399 return(fd); 400} 401 402static int 403byteorder() 404{ 405 u_int32_t x; 406 u_char *p; 407 408 x = 0x01020304; 409 p = (u_char *)&x; 410 switch (*p) { 411 case 1: 412 return (BIG_ENDIAN); 413 case 4: 414 return (LITTLE_ENDIAN); 415 default: 416 return (0); 417 } 418} 419 420int 421__bt_fd(dbp) 422 const DB *dbp; 423{ 424 BTREE *t; 425 426 t = dbp->internal; 427 428 /* Toss any page pinned across calls. */ 429 if (t->bt_pinned != NULL) { 430 mpool_put(t->bt_mp, t->bt_pinned, 0); 431 t->bt_pinned = NULL; 432 } 433 434 /* In-memory database can't have a file descriptor. */ 435 if (ISSET(t, B_INMEM)) { 436 errno = ENOENT; 437 return (-1); 438 } 439 return (t->bt_fd); 440} 441