bt_open.c revision 92986
1130394Sdwmalone/*- 2204977Simp * Copyright (c) 1990, 1993, 1994 3130394Sdwmalone * The Regents of the University of California. All rights reserved. 4130394Sdwmalone * 5130394Sdwmalone * This code is derived from software contributed to Berkeley by 6130394Sdwmalone * Mike Olson. 7130394Sdwmalone * 8130394Sdwmalone * Redistribution and use in source and binary forms, with or without 9130394Sdwmalone * modification, are permitted provided that the following conditions 10130394Sdwmalone * are met: 11130394Sdwmalone * 1. Redistributions of source code must retain the above copyright 12130394Sdwmalone * notice, this list of conditions and the following disclaimer. 13130394Sdwmalone * 2. Redistributions in binary form must reproduce the above copyright 14130394Sdwmalone * notice, this list of conditions and the following disclaimer in the 15130394Sdwmalone * documentation and/or other materials provided with the distribution. 16130394Sdwmalone * 3. All advertising materials mentioning features or use of this software 17130394Sdwmalone * must display the following acknowledgement: 18130394Sdwmalone * This product includes software developed by the University of 19130394Sdwmalone * California, Berkeley and its contributors. 20130394Sdwmalone * 4. Neither the name of the University nor the names of its contributors 21130394Sdwmalone * may be used to endorse or promote products derived from this software 22130394Sdwmalone * without specific prior written permission. 23130394Sdwmalone * 24130394Sdwmalone * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25130394Sdwmalone * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26130394Sdwmalone * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27130394Sdwmalone * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28130394Sdwmalone * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29130394Sdwmalone * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30130394Sdwmalone * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31130394Sdwmalone * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32290052Sjhb * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33290052Sjhb * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34130394Sdwmalone * SUCH DAMAGE. 35290052Sjhb */ 36130394Sdwmalone 37168569Sdelphij#if defined(LIBC_SCCS) && !defined(lint) 38130394Sdwmalonestatic char sccsid[] = "@(#)bt_open.c 8.10 (Berkeley) 8/17/94"; 39130394Sdwmalone#endif /* LIBC_SCCS and not lint */ 40130394Sdwmalone#include <sys/cdefs.h> 41130394Sdwmalone__FBSDID("$FreeBSD: head/lib/libc/db/btree/bt_open.c 92986 2002-03-22 21:53:29Z obrien $"); 42130394Sdwmalone 43130394Sdwmalone/* 44130394Sdwmalone * Implementation of btree access method for 4.4BSD. 45130394Sdwmalone * 46130394Sdwmalone * The design here was originally based on that of the btree access method 47130394Sdwmalone * used in the Postgres database system at UC Berkeley. This implementation 48130394Sdwmalone * is wholly independent of the Postgres code. 49290052Sjhb */ 50290052Sjhb 51240562Szont#include "namespace.h" 52240005Szont#include <sys/param.h> 53240005Szont#include <sys/stat.h> 54290052Sjhb 55240562Szont#include <errno.h> 56290052Sjhb#include <fcntl.h> 57130394Sdwmalone#include <limits.h> 58240562Szont#include <signal.h> 59290052Sjhb#include <stdio.h> 60240562Szont#include <stdlib.h> 61240005Szont#include <string.h> 62290052Sjhb#include <unistd.h> 63240005Szont#include "un-namespace.h" 64130394Sdwmalone 65240005Szont#include <db.h> 66290052Sjhb#include "btree.h" 67240005Szont 68240005Szont#ifdef DEBUG 69290052Sjhb#undef MINPSIZE 70290052Sjhb#define MINPSIZE 128 71290052Sjhb#endif 72290052Sjhb 73240005Szontstatic int byteorder(void); 74240005Szontstatic int nroot(BTREE *); 75290052Sjhbstatic int tmp(void); 76240005Szont 77240005Szont/* 78240005Szont * __BT_OPEN -- Open a btree. 79240005Szont * 80240005Szont * Creates and fills a DB struct, and calls the routine that actually 81130394Sdwmalone * opens the btree. 82290052Sjhb * 83240005Szont * Parameters: 84290052Sjhb * fname: filename (NULL for in-memory trees) 85290052Sjhb * flags: open flag bits 86290052Sjhb * mode: open permission bits 87290052Sjhb * b: BTREEINFO pointer 88290052Sjhb * 89290052Sjhb * Returns: 90240005Szont * NULL on failure, pointer to DB on success. 91240005Szont * 92290052Sjhb */ 93240005SzontDB * 94240005Szont__bt_open(fname, flags, mode, openinfo, dflags) 95290052Sjhb const char *fname; 96290052Sjhb int flags, mode, dflags; 97240562Szont const BTREEINFO *openinfo; 98240005Szont{ 99290052Sjhb struct stat sb; 100240005Szont BTMETA m; 101240005Szont BTREE *t; 102290052Sjhb BTREEINFO b; 103130394Sdwmalone DB *dbp; 104130394Sdwmalone pgno_t ncache; 105290052Sjhb ssize_t nr; 106290052Sjhb int machine_lorder; 107130394Sdwmalone 108240005Szont t = NULL; 109240562Szont 110130394Sdwmalone /* 111240562Szont * Intention is to make sure all of the user's selections are okay 112240562Szont * here and then use them without checking. Can't be complete, since 113240005Szont * we don't know the right page size, lorder or flags until the backing 114240005Szont * file is opened. Also, the file's page size can cause the cachesize 115240005Szont * to change. 116130394Sdwmalone */ 117290052Sjhb machine_lorder = byteorder(); 118290052Sjhb if (openinfo) { 119290052Sjhb b = *openinfo; 120290052Sjhb 121290052Sjhb /* Flags: R_DUP. */ 122130394Sdwmalone if (b.flags & ~(R_DUP)) 123290052Sjhb goto einval; 124290052Sjhb 125290052Sjhb /* 126290052Sjhb * Page size must be indx_t aligned and >= MINPSIZE. Default 127290052Sjhb * page size is set farther on, based on the underlying file 128290052Sjhb * transfer size. 129290052Sjhb */ 130130394Sdwmalone if (b.psize && 131290052Sjhb (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 132 b.psize & (sizeof(indx_t) - 1) )) 133 goto einval; 134 135 /* Minimum number of keys per page; absolute minimum is 2. */ 136 if (b.minkeypage) { 137 if (b.minkeypage < 2) 138 goto einval; 139 } else 140 b.minkeypage = DEFMINKEYPAGE; 141 142 /* If no comparison, use default comparison and prefix. */ 143 if (b.compare == NULL) { 144 b.compare = __bt_defcmp; 145 if (b.prefix == NULL) 146 b.prefix = __bt_defpfx; 147 } 148 149 if (b.lorder == 0) 150 b.lorder = machine_lorder; 151 } else { 152 b.compare = __bt_defcmp; 153 b.cachesize = 0; 154 b.flags = 0; 155 b.lorder = machine_lorder; 156 b.minkeypage = DEFMINKEYPAGE; 157 b.prefix = __bt_defpfx; 158 b.psize = 0; 159 } 160 161 /* Check for the ubiquitous PDP-11. */ 162 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 163 goto einval; 164 165 /* Allocate and initialize DB and BTREE structures. */ 166 if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL) 167 goto err; 168 memset(t, 0, sizeof(BTREE)); 169 t->bt_fd = -1; /* Don't close unopened fd on error. */ 170 t->bt_lorder = b.lorder; 171 t->bt_order = NOT; 172 t->bt_cmp = b.compare; 173 t->bt_pfx = b.prefix; 174 t->bt_rfd = -1; 175 176 if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL) 177 goto err; 178 memset(t->bt_dbp, 0, sizeof(DB)); 179 if (t->bt_lorder != machine_lorder) 180 F_SET(t, B_NEEDSWAP); 181 182 dbp->type = DB_BTREE; 183 dbp->internal = t; 184 dbp->close = __bt_close; 185 dbp->del = __bt_delete; 186 dbp->fd = __bt_fd; 187 dbp->get = __bt_get; 188 dbp->put = __bt_put; 189 dbp->seq = __bt_seq; 190 dbp->sync = __bt_sync; 191 192 /* 193 * If no file name was supplied, this is an in-memory btree and we 194 * open a backing temporary file. Otherwise, it's a disk-based tree. 195 */ 196 if (fname) { 197 switch (flags & O_ACCMODE) { 198 case O_RDONLY: 199 F_SET(t, B_RDONLY); 200 break; 201 case O_RDWR: 202 break; 203 case O_WRONLY: 204 default: 205 goto einval; 206 } 207 208 if ((t->bt_fd = _open(fname, flags, mode)) < 0) 209 goto err; 210 211 } else { 212 if ((flags & O_ACCMODE) != O_RDWR) 213 goto einval; 214 if ((t->bt_fd = tmp()) == -1) 215 goto err; 216 F_SET(t, B_INMEM); 217 } 218 219 if (_fcntl(t->bt_fd, F_SETFD, 1) == -1) 220 goto err; 221 222 if (_fstat(t->bt_fd, &sb)) 223 goto err; 224 if (sb.st_size) { 225 if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 226 goto err; 227 if (nr != sizeof(BTMETA)) 228 goto eftype; 229 230 /* 231 * Read in the meta-data. This can change the notion of what 232 * the lorder, page size and flags are, and, when the page size 233 * changes, the cachesize value can change too. If the user 234 * specified the wrong byte order for an existing database, we 235 * don't bother to return an error, we just clear the NEEDSWAP 236 * bit. 237 */ 238 if (m.magic == BTREEMAGIC) 239 F_CLR(t, B_NEEDSWAP); 240 else { 241 F_SET(t, B_NEEDSWAP); 242 M_32_SWAP(m.magic); 243 M_32_SWAP(m.version); 244 M_32_SWAP(m.psize); 245 M_32_SWAP(m.free); 246 M_32_SWAP(m.nrecs); 247 M_32_SWAP(m.flags); 248 } 249 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 250 goto eftype; 251 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 252 m.psize & (sizeof(indx_t) - 1) ) 253 goto eftype; 254 if (m.flags & ~SAVEMETA) 255 goto eftype; 256 b.psize = m.psize; 257 F_SET(t, m.flags); 258 t->bt_free = m.free; 259 t->bt_nrecs = m.nrecs; 260 } else { 261 /* 262 * Set the page size to the best value for I/O to this file. 263 * Don't overflow the page offset type. 264 */ 265 if (b.psize == 0) { 266 b.psize = sb.st_blksize; 267 if (b.psize < MINPSIZE) 268 b.psize = MINPSIZE; 269 if (b.psize > MAX_PAGE_OFFSET + 1) 270 b.psize = MAX_PAGE_OFFSET + 1; 271 } 272 273 /* Set flag if duplicates permitted. */ 274 if (!(b.flags & R_DUP)) 275 F_SET(t, B_NODUPS); 276 277 t->bt_free = P_INVALID; 278 t->bt_nrecs = 0; 279 F_SET(t, B_METADIRTY); 280 } 281 282 t->bt_psize = b.psize; 283 284 /* Set the cache size; must be a multiple of the page size. */ 285 if (b.cachesize && b.cachesize & (b.psize - 1) ) 286 b.cachesize += (~b.cachesize & (b.psize - 1) ) + 1; 287 if (b.cachesize < b.psize * MINCACHE) 288 b.cachesize = b.psize * MINCACHE; 289 290 /* Calculate number of pages to cache. */ 291 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 292 293 /* 294 * The btree data structure requires that at least two keys can fit on 295 * a page, but other than that there's no fixed requirement. The user 296 * specified a minimum number per page, and we translated that into the 297 * number of bytes a key/data pair can use before being placed on an 298 * overflow page. This calculation includes the page header, the size 299 * of the index referencing the leaf item and the size of the leaf item 300 * structure. Also, don't let the user specify a minkeypage such that 301 * a key/data pair won't fit even if both key and data are on overflow 302 * pages. 303 */ 304 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 305 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 306 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 307 t->bt_ovflsize = 308 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 309 310 /* Initialize the buffer pool. */ 311 if ((t->bt_mp = 312 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 313 goto err; 314 if (!F_ISSET(t, B_INMEM)) 315 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 316 317 /* Create a root page if new tree. */ 318 if (nroot(t) == RET_ERROR) 319 goto err; 320 321 /* Global flags. */ 322 if (dflags & DB_LOCK) 323 F_SET(t, B_DB_LOCK); 324 if (dflags & DB_SHMEM) 325 F_SET(t, B_DB_SHMEM); 326 if (dflags & DB_TXN) 327 F_SET(t, B_DB_TXN); 328 329 return (dbp); 330 331einval: errno = EINVAL; 332 goto err; 333 334eftype: errno = EFTYPE; 335 goto err; 336 337err: if (t) { 338 if (t->bt_dbp) 339 free(t->bt_dbp); 340 if (t->bt_fd != -1) 341 (void)_close(t->bt_fd); 342 free(t); 343 } 344 return (NULL); 345} 346 347/* 348 * NROOT -- Create the root of a new tree. 349 * 350 * Parameters: 351 * t: tree 352 * 353 * Returns: 354 * RET_ERROR, RET_SUCCESS 355 */ 356static int 357nroot(t) 358 BTREE *t; 359{ 360 PAGE *meta, *root; 361 pgno_t npg; 362 363 if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { 364 mpool_put(t->bt_mp, meta, 0); 365 return (RET_SUCCESS); 366 } 367 if (errno != EINVAL) /* It's OK to not exist. */ 368 return (RET_ERROR); 369 errno = 0; 370 371 if ((meta = mpool_new(t->bt_mp, &npg)) == NULL) 372 return (RET_ERROR); 373 374 if ((root = mpool_new(t->bt_mp, &npg)) == NULL) 375 return (RET_ERROR); 376 377 if (npg != P_ROOT) 378 return (RET_ERROR); 379 root->pgno = npg; 380 root->prevpg = root->nextpg = P_INVALID; 381 root->lower = BTDATAOFF; 382 root->upper = t->bt_psize; 383 root->flags = P_BLEAF; 384 memset(meta, 0, t->bt_psize); 385 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 386 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 387 return (RET_SUCCESS); 388} 389 390static int 391tmp() 392{ 393 sigset_t set, oset; 394 int fd; 395 char *envtmp = NULL; 396 char path[MAXPATHLEN]; 397 398 if (issetugid() == 0) 399 envtmp = getenv("TMPDIR"); 400 (void)snprintf(path, 401 sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp"); 402 403 (void)sigfillset(&set); 404 (void)_sigprocmask(SIG_BLOCK, &set, &oset); 405 if ((fd = mkstemp(path)) != -1) 406 (void)unlink(path); 407 (void)_sigprocmask(SIG_SETMASK, &oset, NULL); 408 return(fd); 409} 410 411static int 412byteorder() 413{ 414 u_int32_t x; 415 u_char *p; 416 417 x = 0x01020304; 418 p = (u_char *)&x; 419 switch (*p) { 420 case 1: 421 return (BIG_ENDIAN); 422 case 4: 423 return (LITTLE_ENDIAN); 424 default: 425 return (0); 426 } 427} 428 429int 430__bt_fd(dbp) 431 const DB *dbp; 432{ 433 BTREE *t; 434 435 t = dbp->internal; 436 437 /* Toss any page pinned across calls. */ 438 if (t->bt_pinned != NULL) { 439 mpool_put(t->bt_mp, t->bt_pinned, 0); 440 t->bt_pinned = NULL; 441 } 442 443 /* In-memory database can't have a file descriptor. */ 444 if (F_ISSET(t, B_INMEM)) { 445 errno = ENOENT; 446 return (-1); 447 } 448 return (t->bt_fd); 449} 450