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 * @(#)bt_open.c 8.10 (Berkeley) 8/17/94 33 * $FreeBSD: head/lib/libc/db/btree/bt_open.c 190498 2009-03-28 07:31:02Z delphij $ 34 */ 35 36 /* 37 * Implementation of btree access method for 4.4BSD. 38 * 39 * The design here was originally based on that of the btree access method 40 * used in the Postgres database system at UC Berkeley. This implementation 41 * is wholly independent of the Postgres code. 42 */ 43 44 #include "namespace.h" 45 #include <sys/param.h> 46 #include <sys/stat.h> 47 #include <sys/file.h> 48 49 #include <errno.h> 50 #include <fcntl.h> 51 #include <limits.h> 52 #include <signal.h> 53 #include <stdio.h> 54 #include <stdlib.h> 55 #include <string.h> 56 #include <unistd.h> 57 #include "un-namespace.h" 58 59 #include <db.h> 60 #include "btree.h" 61 62 #ifdef DEBUG 63 #undef MINPSIZE 64 #define MINPSIZE 128 65 #endif 66 67 static int byteorder(void); 68 static int nroot(BTREE *); 69 static int tmp(void); 70 71 /* 72 * __BT_OPEN -- Open a btree. 73 * 74 * Creates and fills a DB struct, and calls the routine that actually 75 * opens the btree. 76 * 77 * Parameters: 78 * fname: filename (NULL for in-memory trees) 79 * flags: open flag bits 80 * mode: open permission bits 81 * b: BTREEINFO pointer 82 * 83 * Returns: 84 * NULL on failure, pointer to DB on success. 85 * 86 */ 87 DB * 88 __bt_open(const char *fname, int flags, mode_t mode, const BTREEINFO *openinfo, 89 int dflags) 90 { 91 struct stat sb; 92 BTMETA m; 93 BTREE *t; 94 BTREEINFO b; 95 DB *dbp; 96 pgno_t ncache; 97 ssize_t nr; 98 int machine_lorder, saved_errno; 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 = (BTREE *)calloc(1, sizeof(BTREE))) == NULL) 159 goto err; 160 t->bt_fd = -1; /* Don't close unopened fd on error. */ 161 t->bt_lorder = b.lorder; 162 t->bt_order = NOT; 163 t->bt_cmp = b.compare; 164 t->bt_pfx = b.prefix; 165 t->bt_rfd = -1; 166 167 if ((t->bt_dbp = dbp = (DB *)calloc(1, sizeof(DB))) == NULL) 168 goto err; 169 if (t->bt_lorder != machine_lorder) 170 F_SET(t, B_NEEDSWAP); 171 172 dbp->type = DB_BTREE; 173 dbp->internal = t; 174 dbp->close = __bt_close; 175 dbp->del = __bt_delete; 176 dbp->fd = __bt_fd; 177 dbp->get = __bt_get; 178 dbp->put = __bt_put; 179 dbp->seq = __bt_seq; 180 dbp->sync = __bt_sync; 181 182 /* 183 * If no file name was supplied, this is an in-memory btree and we 184 * open a backing temporary file. Otherwise, it's a disk-based tree. 185 */ 186 if (fname) { 187 switch (flags & O_ACCMODE) { 188 case O_RDONLY: 189 F_SET(t, B_RDONLY); 190 break; 191 case O_RDWR: 192 break; 193 case O_WRONLY: 194 default: 195 goto einval; 196 } 197 198 if ((t->bt_fd = _open(fname, flags | O_CLOEXEC, mode)) < 0) 199 goto err; 200 201 } else { 202 if ((flags & O_ACCMODE) != O_RDWR) 203 goto einval; 204 if ((t->bt_fd = tmp()) == -1) 205 goto err; 206 F_SET(t, B_INMEM); 207 } 208 209 if (_fcntl(t->bt_fd, F_SETFD, 1) == -1) 210 goto err; 211 212 if (_fstat(t->bt_fd, &sb)) 213 goto err; 214 if (sb.st_size) { 215 if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 216 goto err; 217 if (nr != sizeof(BTMETA)) 218 goto eftype; 219 220 /* 221 * Read in the meta-data. This can change the notion of what 222 * the lorder, page size and flags are, and, when the page size 223 * changes, the cachesize value can change too. If the user 224 * specified the wrong byte order for an existing database, we 225 * don't bother to return an error, we just clear the NEEDSWAP 226 * bit. 227 */ 228 if (m.magic == BTREEMAGIC) 229 F_CLR(t, B_NEEDSWAP); 230 else { 231 F_SET(t, B_NEEDSWAP); 232 M_32_SWAP(m.magic); 233 M_32_SWAP(m.version); 234 M_32_SWAP(m.psize); 235 M_32_SWAP(m.free); 236 M_32_SWAP(m.nrecs); 237 M_32_SWAP(m.flags); 238 } 239 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 240 goto eftype; 241 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 242 m.psize & (sizeof(indx_t) - 1) ) 243 goto eftype; 244 if (m.flags & ~SAVEMETA) 245 goto eftype; 246 b.psize = m.psize; 247 F_SET(t, m.flags); 248 t->bt_free = m.free; 249 t->bt_nrecs = m.nrecs; 250 } else { 251 /* 252 * Set the page size to the best value for I/O to this file. 253 * Don't overflow the page offset type. 254 * 255 * Stop using st_blksize, its meaningless on a modern system 256 * and can cause db to operate inefficiently. Instead we 257 * use NOMPSIZE. 258 */ 259 if (b.psize == 0) { 260 b.psize = NOMPSIZE; 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 F_SET(t, B_NODUPS); 270 271 t->bt_free = P_INVALID; 272 t->bt_nrecs = 0; 273 F_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 (!F_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 F_SET(t, B_DB_LOCK); 318 if (dflags & DB_SHMEM) 319 F_SET(t, B_DB_SHMEM); 320 if (dflags & DB_TXN) 321 F_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: saved_errno = errno; 332 if (t) { 333 if (t->bt_dbp) 334 free(t->bt_dbp); 335 if (t->bt_fd != -1) 336 _close(t->bt_fd); 337 free(t); 338 } 339 errno = saved_errno; 340 return (NULL); 341 } 342 343 /* 344 * NROOT -- Create the root of a new tree. 345 * 346 * Parameters: 347 * t: tree 348 * 349 * Returns: 350 * RET_ERROR, RET_SUCCESS 351 */ 352 static int 353 nroot(BTREE *t) 354 { 355 PAGE *meta, *root; 356 pgno_t npg; 357 358 if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) { 359 if (root->lower == 0 && 360 root->pgno == 0 && 361 root->linp[0] == 0) { 362 mpool_delete(t->bt_mp, root); 363 errno = EINVAL; 364 } else { 365 mpool_put(t->bt_mp, root, 0); 366 return (RET_SUCCESS); 367 } 368 } 369 if (errno != EINVAL) /* It's OK to not exist. */ 370 return (RET_ERROR); 371 errno = 0; 372 373 if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL) 374 return (RET_ERROR); 375 376 if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL) 377 return (RET_ERROR); 378 379 if (npg != P_ROOT) 380 return (RET_ERROR); 381 root->pgno = npg; 382 root->prevpg = root->nextpg = P_INVALID; 383 root->lower = BTDATAOFF; 384 root->upper = t->bt_psize; 385 root->flags = P_BLEAF; 386 memset(meta, 0, t->bt_psize); 387 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 388 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 389 return (RET_SUCCESS); 390 } 391 392 static int 393 tmp(void) 394 { 395 sigset_t set, oset; 396 int fd, len; 397 char *envtmp = NULL; 398 char path[MAXPATHLEN]; 399 400 if (issetugid() == 0) 401 envtmp = getenv("TMPDIR"); 402 len = snprintf(path, 403 sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp"); 404 if (len < 0 || len >= (int)sizeof(path)) { 405 errno = ENAMETOOLONG; 406 return(-1); 407 } 408 409 sigfillset(&set); 410 _sigprocmask(SIG_BLOCK, &set, &oset); 411 if ((fd = mkostemp(path, O_CLOEXEC)) != -1) 412 _unlink(path); 413 _sigprocmask(SIG_SETMASK, &oset, NULL); 414 return(fd); 415 } 416 417 static int 418 byteorder(void) 419 { 420 uint32_t x; 421 unsigned char *p; 422 423 x = 0x01020304; 424 p = (unsigned char *)&x; 425 switch (*p) { 426 case 1: 427 return (BIG_ENDIAN); 428 case 4: 429 return (LITTLE_ENDIAN); 430 default: 431 return (0); 432 } 433 } 434 435 int 436 __bt_fd(const DB *dbp) 437 { 438 BTREE *t; 439 440 t = dbp->internal; 441 442 /* Toss any page pinned across calls. */ 443 if (t->bt_pinned != NULL) { 444 mpool_put(t->bt_mp, t->bt_pinned, 0); 445 t->bt_pinned = NULL; 446 } 447 448 /* In-memory database can't have a file descriptor. */ 449 if (F_ISSET(t, B_INMEM)) { 450 errno = ENOENT; 451 return (-1); 452 } 453 return (t->bt_fd); 454 } 455