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. Neither the name of the University nor the names of its contributors 17 * may be used to endorse or promote products derived from this software 18 * without specific prior written permission. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 * 32 * $FreeBSD: src/lib/libc/db/btree/bt_open.c,v 1.7.2.1 2000/11/02 10:30:07 kris Exp $ 33 * $DragonFly: src/lib/libc/db/btree/bt_open.c,v 1.5 2005/09/19 09:20:37 asmodai Exp $ 34 * 35 * @(#)bt_open.c 8.10 (Berkeley) 8/17/94 36 */ 37 38 /* 39 * Implementation of btree access method for 4.4BSD. 40 * 41 * The design here was originally based on that of the btree access method 42 * used in the Postgres database system at UC Berkeley. This implementation 43 * is wholly independent of the Postgres code. 44 */ 45 46 #include "namespace.h" 47 #include <sys/param.h> 48 #include <sys/stat.h> 49 50 #include <errno.h> 51 #include <fcntl.h> 52 #include <limits.h> 53 #include <signal.h> 54 #include <stdio.h> 55 #include <stdlib.h> 56 #include <string.h> 57 #include <unistd.h> 58 #include "un-namespace.h" 59 60 #include <db.h> 61 #include "btree.h" 62 63 #ifdef DEBUG 64 #undef MINPSIZE 65 #define MINPSIZE 128 66 #endif 67 68 static int byteorder (void); 69 static int nroot (BTREE *); 70 static int tmp (void); 71 72 /* 73 * __BT_OPEN -- Open a btree. 74 * 75 * Creates and fills a DB struct, and calls the routine that actually 76 * opens the btree. 77 * 78 * Parameters: 79 * fname: filename (NULL for in-memory trees) 80 * flags: open flag bits 81 * mode: open permission bits 82 * b: BTREEINFO pointer 83 * 84 * Returns: 85 * NULL on failure, pointer to DB on success. 86 * 87 */ 88 DB * 89 __bt_open(fname, flags, mode, openinfo, dflags) 90 const char *fname; 91 int flags, mode, dflags; 92 const BTREEINFO *openinfo; 93 { 94 struct stat sb; 95 BTMETA m; 96 BTREE *t; 97 BTREEINFO b; 98 DB *dbp; 99 pgno_t ncache; 100 ssize_t nr; 101 int machine_lorder; 102 103 t = NULL; 104 105 /* 106 * Intention is to make sure all of the user's selections are okay 107 * here and then use them without checking. Can't be complete, since 108 * we don't know the right page size, lorder or flags until the backing 109 * file is opened. Also, the file's page size can cause the cachesize 110 * to change. 111 */ 112 machine_lorder = byteorder(); 113 if (openinfo) { 114 b = *openinfo; 115 116 /* Flags: R_DUP. */ 117 if (b.flags & ~(R_DUP)) 118 goto einval; 119 120 /* 121 * Page size must be indx_t aligned and >= MINPSIZE. Default 122 * page size is set farther on, based on the underlying file 123 * transfer size. 124 */ 125 if (b.psize && 126 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 127 b.psize & (sizeof(indx_t) - 1) )) 128 goto einval; 129 130 /* Minimum number of keys per page; absolute minimum is 2. */ 131 if (b.minkeypage) { 132 if (b.minkeypage < 2) 133 goto einval; 134 } else 135 b.minkeypage = DEFMINKEYPAGE; 136 137 /* If no comparison, use default comparison and prefix. */ 138 if (b.compare == NULL) { 139 b.compare = __bt_defcmp; 140 if (b.prefix == NULL) 141 b.prefix = __bt_defpfx; 142 } 143 144 if (b.lorder == 0) 145 b.lorder = machine_lorder; 146 } else { 147 b.compare = __bt_defcmp; 148 b.cachesize = 0; 149 b.flags = 0; 150 b.lorder = machine_lorder; 151 b.minkeypage = DEFMINKEYPAGE; 152 b.prefix = __bt_defpfx; 153 b.psize = 0; 154 } 155 156 /* Check for the ubiquitous PDP-11. */ 157 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 158 goto einval; 159 160 /* Allocate and initialize DB and BTREE structures. */ 161 if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL) 162 goto err; 163 memset(t, 0, sizeof(BTREE)); 164 t->bt_fd = -1; /* Don't close unopened fd on error. */ 165 t->bt_lorder = b.lorder; 166 t->bt_order = NOT; 167 t->bt_cmp = b.compare; 168 t->bt_pfx = b.prefix; 169 t->bt_rfd = -1; 170 171 if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL) 172 goto err; 173 memset(t->bt_dbp, 0, sizeof(DB)); 174 if (t->bt_lorder != machine_lorder) 175 F_SET(t, B_NEEDSWAP); 176 177 dbp->type = DB_BTREE; 178 dbp->internal = t; 179 dbp->close = __bt_close; 180 dbp->del = __bt_delete; 181 dbp->fd = __bt_fd; 182 dbp->get = __bt_get; 183 dbp->put = __bt_put; 184 dbp->seq = __bt_seq; 185 dbp->sync = __bt_sync; 186 187 /* 188 * If no file name was supplied, this is an in-memory btree and we 189 * open a backing temporary file. Otherwise, it's a disk-based tree. 190 */ 191 if (fname) { 192 switch (flags & O_ACCMODE) { 193 case O_RDONLY: 194 F_SET(t, B_RDONLY); 195 break; 196 case O_RDWR: 197 break; 198 case O_WRONLY: 199 default: 200 goto einval; 201 } 202 203 if ((t->bt_fd = _open(fname, 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 F_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 if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 221 goto err; 222 if (nr != sizeof(BTMETA)) 223 goto eftype; 224 225 /* 226 * Read in the meta-data. This can change the notion of what 227 * the lorder, page size and flags are, and, when the page size 228 * changes, the cachesize value can change too. If the user 229 * specified the wrong byte order for an existing database, we 230 * don't bother to return an error, we just clear the NEEDSWAP 231 * bit. 232 */ 233 if (m.magic == BTREEMAGIC) 234 F_CLR(t, B_NEEDSWAP); 235 else { 236 F_SET(t, B_NEEDSWAP); 237 M_32_SWAP(m.magic); 238 M_32_SWAP(m.version); 239 M_32_SWAP(m.psize); 240 M_32_SWAP(m.free); 241 M_32_SWAP(m.nrecs); 242 M_32_SWAP(m.flags); 243 } 244 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 245 goto eftype; 246 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 247 m.psize & (sizeof(indx_t) - 1) ) 248 goto eftype; 249 if (m.flags & ~SAVEMETA) 250 goto eftype; 251 b.psize = m.psize; 252 F_SET(t, m.flags); 253 t->bt_free = m.free; 254 t->bt_nrecs = m.nrecs; 255 } else { 256 /* 257 * Set the page size to the best value for I/O to this file. 258 * Don't overflow the page offset type. 259 */ 260 if (b.psize == 0) { 261 b.psize = sb.st_blksize; 262 if (b.psize < MINPSIZE) 263 b.psize = MINPSIZE; 264 if (b.psize > MAX_PAGE_OFFSET + 1) 265 b.psize = MAX_PAGE_OFFSET + 1; 266 } 267 268 /* Set flag if duplicates permitted. */ 269 if (!(b.flags & R_DUP)) 270 F_SET(t, B_NODUPS); 271 272 t->bt_free = P_INVALID; 273 t->bt_nrecs = 0; 274 F_SET(t, B_METADIRTY); 275 } 276 277 t->bt_psize = b.psize; 278 279 /* Set the cache size; must be a multiple of the page size. */ 280 if (b.cachesize && b.cachesize & (b.psize - 1) ) 281 b.cachesize += (~b.cachesize & (b.psize - 1) ) + 1; 282 if (b.cachesize < b.psize * MINCACHE) 283 b.cachesize = b.psize * MINCACHE; 284 285 /* Calculate number of pages to cache. */ 286 ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize; 287 288 /* 289 * The btree data structure requires that at least two keys can fit on 290 * a page, but other than that there's no fixed requirement. The user 291 * specified a minimum number per page, and we translated that into the 292 * number of bytes a key/data pair can use before being placed on an 293 * overflow page. This calculation includes the page header, the size 294 * of the index referencing the leaf item and the size of the leaf item 295 * structure. Also, don't let the user specify a minkeypage such that 296 * a key/data pair won't fit even if both key and data are on overflow 297 * pages. 298 */ 299 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 300 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 301 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 302 t->bt_ovflsize = 303 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 304 305 /* Initialize the buffer pool. */ 306 if ((t->bt_mp = 307 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 308 goto err; 309 if (!F_ISSET(t, B_INMEM)) 310 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 311 312 /* Create a root page if new tree. */ 313 if (nroot(t) == RET_ERROR) 314 goto err; 315 316 /* Global flags. */ 317 if (dflags & DB_LOCK) 318 F_SET(t, B_DB_LOCK); 319 if (dflags & DB_SHMEM) 320 F_SET(t, B_DB_SHMEM); 321 if (dflags & DB_TXN) 322 F_SET(t, B_DB_TXN); 323 324 return (dbp); 325 326 einval: errno = EINVAL; 327 goto err; 328 329 eftype: errno = EFTYPE; 330 goto err; 331 332 err: if (t) { 333 if (t->bt_dbp) 334 free(t->bt_dbp); 335 if (t->bt_fd != -1) 336 (void)_close(t->bt_fd); 337 free(t); 338 } 339 return (NULL); 340 } 341 342 /* 343 * NROOT -- Create the root of a new tree. 344 * 345 * Parameters: 346 * t: tree 347 * 348 * Returns: 349 * RET_ERROR, RET_SUCCESS 350 */ 351 static int 352 nroot(t) 353 BTREE *t; 354 { 355 PAGE *meta, *root; 356 pgno_t npg; 357 358 if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) { 359 mpool_put(t->bt_mp, meta, 0); 360 return (RET_SUCCESS); 361 } 362 if (errno != EINVAL) /* It's OK to not exist. */ 363 return (RET_ERROR); 364 errno = 0; 365 366 if ((meta = mpool_new(t->bt_mp, &npg)) == NULL) 367 return (RET_ERROR); 368 369 if ((root = mpool_new(t->bt_mp, &npg)) == NULL) 370 return (RET_ERROR); 371 372 if (npg != P_ROOT) 373 return (RET_ERROR); 374 root->pgno = npg; 375 root->prevpg = root->nextpg = P_INVALID; 376 root->lower = BTDATAOFF; 377 root->upper = t->bt_psize; 378 root->flags = P_BLEAF; 379 memset(meta, 0, t->bt_psize); 380 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 381 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 382 return (RET_SUCCESS); 383 } 384 385 static int 386 tmp() 387 { 388 sigset_t set, oset; 389 int fd; 390 char *envtmp = NULL; 391 char path[MAXPATHLEN]; 392 393 if (issetugid() == 0) 394 envtmp = getenv("TMPDIR"); 395 (void)snprintf(path, 396 sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp"); 397 398 (void)sigfillset(&set); 399 (void)_sigprocmask(SIG_BLOCK, &set, &oset); 400 if ((fd = mkstemp(path)) != -1) 401 (void)unlink(path); 402 (void)_sigprocmask(SIG_SETMASK, &oset, NULL); 403 return(fd); 404 } 405 406 static int 407 byteorder() 408 { 409 u_int32_t x; 410 u_char *p; 411 412 x = 0x01020304; 413 p = (u_char *)&x; 414 switch (*p) { 415 case 1: 416 return (BIG_ENDIAN); 417 case 4: 418 return (LITTLE_ENDIAN); 419 default: 420 return (0); 421 } 422 } 423 424 int 425 __bt_fd(dbp) 426 const DB *dbp; 427 { 428 BTREE *t; 429 430 t = dbp->internal; 431 432 /* Toss any page pinned across calls. */ 433 if (t->bt_pinned != NULL) { 434 mpool_put(t->bt_mp, t->bt_pinned, 0); 435 t->bt_pinned = NULL; 436 } 437 438 /* In-memory database can't have a file descriptor. */ 439 if (F_ISSET(t, B_INMEM)) { 440 errno = ENOENT; 441 return (-1); 442 } 443 return (t->bt_fd); 444 } 445