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 33 #if defined(LIBC_SCCS) && !defined(lint) 34 static char sccsid[] = "@(#)bt_open.c 8.10 (Berkeley) 8/17/94"; 35 #endif /* LIBC_SCCS and not lint */ 36 #include <sys/cdefs.h> 37 __FBSDID("$FreeBSD$"); 38 39 /* 40 * Implementation of btree access method for 4.4BSD. 41 * 42 * The design here was originally based on that of the btree access method 43 * used in the Postgres database system at UC Berkeley. This implementation 44 * is wholly independent of the Postgres code. 45 */ 46 47 #include "namespace.h" 48 #include <sys/param.h> 49 #include <sys/stat.h> 50 51 #include <errno.h> 52 #include <fcntl.h> 53 #include <limits.h> 54 #include <signal.h> 55 #include <stdio.h> 56 #include <stdlib.h> 57 #include <string.h> 58 #include <unistd.h> 59 #include "un-namespace.h" 60 #include "libc_private.h" 61 62 #include <db.h> 63 #include "btree.h" 64 65 #ifdef DEBUG 66 #undef MINPSIZE 67 #define MINPSIZE 128 68 #endif 69 70 static int byteorder(void); 71 static int nroot(BTREE *); 72 static int tmp(void); 73 74 /* 75 * __BT_OPEN -- Open a btree. 76 * 77 * Creates and fills a DB struct, and calls the routine that actually 78 * opens the btree. 79 * 80 * Parameters: 81 * fname: filename (NULL for in-memory trees) 82 * flags: open flag bits 83 * mode: open permission bits 84 * b: BTREEINFO pointer 85 * 86 * Returns: 87 * NULL on failure, pointer to DB on success. 88 * 89 */ 90 DB * 91 __bt_open(const char *fname, int flags, int mode, const BTREEINFO *openinfo, int dflags) 92 { 93 struct stat sb; 94 BTMETA m; 95 BTREE *t; 96 BTREEINFO b; 97 DB *dbp; 98 pgno_t ncache; 99 ssize_t nr; 100 int machine_lorder, saved_errno; 101 102 t = NULL; 103 104 /* 105 * Intention is to make sure all of the user's selections are okay 106 * here and then use them without checking. Can't be complete, since 107 * we don't know the right page size, lorder or flags until the backing 108 * file is opened. Also, the file's page size can cause the cachesize 109 * to change. 110 */ 111 machine_lorder = byteorder(); 112 if (openinfo) { 113 b = *openinfo; 114 115 /* Flags: R_DUP. */ 116 if (b.flags & ~(R_DUP)) 117 goto einval; 118 119 /* 120 * Page size must be indx_t aligned and >= MINPSIZE. Default 121 * page size is set farther on, based on the underlying file 122 * transfer size. 123 */ 124 if (b.psize && 125 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 126 b.psize & (sizeof(indx_t) - 1) )) 127 goto einval; 128 129 /* Minimum number of keys per page; absolute minimum is 2. */ 130 if (b.minkeypage) { 131 if (b.minkeypage < 2) 132 goto einval; 133 } else 134 b.minkeypage = DEFMINKEYPAGE; 135 136 /* If no comparison, use default comparison and prefix. */ 137 if (b.compare == NULL) { 138 b.compare = __bt_defcmp; 139 if (b.prefix == NULL) 140 b.prefix = __bt_defpfx; 141 } 142 143 if (b.lorder == 0) 144 b.lorder = machine_lorder; 145 } else { 146 b.compare = __bt_defcmp; 147 b.cachesize = 0; 148 b.flags = 0; 149 b.lorder = machine_lorder; 150 b.minkeypage = DEFMINKEYPAGE; 151 b.prefix = __bt_defpfx; 152 b.psize = 0; 153 } 154 155 /* Check for the ubiquitous PDP-11. */ 156 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 157 goto einval; 158 159 /* Allocate and initialize DB and BTREE structures. */ 160 if ((t = (BTREE *)calloc(1, sizeof(BTREE))) == NULL) 161 goto err; 162 t->bt_fd = -1; /* Don't close unopened fd on error. */ 163 t->bt_lorder = b.lorder; 164 t->bt_order = NOT; 165 t->bt_cmp = b.compare; 166 t->bt_pfx = b.prefix; 167 t->bt_rfd = -1; 168 169 if ((t->bt_dbp = dbp = (DB *)calloc(1, sizeof(DB))) == NULL) 170 goto err; 171 if (t->bt_lorder != machine_lorder) 172 F_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 F_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 | O_CLOEXEC, 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 F_SET(t, B_INMEM); 209 } 210 211 if (_fstat(t->bt_fd, &sb)) 212 goto err; 213 if (sb.st_size) { 214 if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 215 goto err; 216 if (nr != sizeof(BTMETA)) 217 goto eftype; 218 219 /* 220 * Read in the meta-data. This can change the notion of what 221 * the lorder, page size and flags are, and, when the page size 222 * changes, the cachesize value can change too. If the user 223 * specified the wrong byte order for an existing database, we 224 * don't bother to return an error, we just clear the NEEDSWAP 225 * bit. 226 */ 227 if (m.magic == BTREEMAGIC) 228 F_CLR(t, B_NEEDSWAP); 229 else { 230 F_SET(t, B_NEEDSWAP); 231 M_32_SWAP(m.magic); 232 M_32_SWAP(m.version); 233 M_32_SWAP(m.psize); 234 M_32_SWAP(m.free); 235 M_32_SWAP(m.nrecs); 236 M_32_SWAP(m.flags); 237 } 238 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 239 goto eftype; 240 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 241 m.psize & (sizeof(indx_t) - 1) ) 242 goto eftype; 243 if (m.flags & ~SAVEMETA) 244 goto eftype; 245 b.psize = m.psize; 246 F_SET(t, m.flags); 247 t->bt_free = m.free; 248 t->bt_nrecs = m.nrecs; 249 } else { 250 /* 251 * Set the page size to the best value for I/O to this file. 252 * Don't overflow the page offset type. 253 */ 254 if (b.psize == 0) { 255 b.psize = sb.st_blksize; 256 if (b.psize < MINPSIZE) 257 b.psize = MINPSIZE; 258 if (b.psize > MAX_PAGE_OFFSET + 1) 259 b.psize = MAX_PAGE_OFFSET + 1; 260 } 261 262 /* Set flag if duplicates permitted. */ 263 if (!(b.flags & R_DUP)) 264 F_SET(t, B_NODUPS); 265 266 t->bt_free = P_INVALID; 267 t->bt_nrecs = 0; 268 F_SET(t, B_METADIRTY); 269 } 270 271 t->bt_psize = b.psize; 272 273 /* Set the cache size; must be a multiple of the page size. */ 274 if (b.cachesize && b.cachesize & (b.psize - 1) ) 275 b.cachesize += (~b.cachesize & (b.psize - 1) ) + 1; 276 if (b.cachesize < b.psize * MINCACHE) 277 b.cachesize = b.psize * MINCACHE; 278 279 /* Calculate number of pages to cache. */ 280 ncache = howmany(b.cachesize, t->bt_psize); 281 282 /* 283 * The btree data structure requires that at least two keys can fit on 284 * a page, but other than that there's no fixed requirement. The user 285 * specified a minimum number per page, and we translated that into the 286 * number of bytes a key/data pair can use before being placed on an 287 * overflow page. This calculation includes the page header, the size 288 * of the index referencing the leaf item and the size of the leaf item 289 * structure. Also, don't let the user specify a minkeypage such that 290 * a key/data pair won't fit even if both key and data are on overflow 291 * pages. 292 */ 293 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 294 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 295 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 296 t->bt_ovflsize = 297 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 298 299 /* Initialize the buffer pool. */ 300 if ((t->bt_mp = 301 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 302 goto err; 303 if (!F_ISSET(t, B_INMEM)) 304 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 305 306 /* Create a root page if new tree. */ 307 if (nroot(t) == RET_ERROR) 308 goto err; 309 310 /* Global flags. */ 311 if (dflags & DB_LOCK) 312 F_SET(t, B_DB_LOCK); 313 if (dflags & DB_SHMEM) 314 F_SET(t, B_DB_SHMEM); 315 if (dflags & DB_TXN) 316 F_SET(t, B_DB_TXN); 317 318 return (dbp); 319 320 einval: errno = EINVAL; 321 goto err; 322 323 eftype: errno = EFTYPE; 324 goto err; 325 326 err: saved_errno = errno; 327 if (t) { 328 if (t->bt_dbp) 329 free(t->bt_dbp); 330 if (t->bt_fd != -1) 331 (void)_close(t->bt_fd); 332 free(t); 333 } 334 errno = saved_errno; 335 return (NULL); 336 } 337 338 /* 339 * NROOT -- Create the root of a new tree. 340 * 341 * Parameters: 342 * t: tree 343 * 344 * Returns: 345 * RET_ERROR, RET_SUCCESS 346 */ 347 static int 348 nroot(BTREE *t) 349 { 350 PAGE *meta, *root; 351 pgno_t npg; 352 353 if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) { 354 if (root->lower == 0 && 355 root->pgno == 0 && 356 root->linp[0] == 0) { 357 mpool_delete(t->bt_mp, root); 358 errno = EINVAL; 359 } else { 360 mpool_put(t->bt_mp, root, 0); 361 return (RET_SUCCESS); 362 } 363 } 364 if (errno != EINVAL) /* It's OK to not exist. */ 365 return (RET_ERROR); 366 errno = 0; 367 368 if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL) 369 return (RET_ERROR); 370 371 if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL) 372 return (RET_ERROR); 373 374 if (npg != P_ROOT) 375 return (RET_ERROR); 376 root->pgno = npg; 377 root->prevpg = root->nextpg = P_INVALID; 378 root->lower = BTDATAOFF; 379 root->upper = t->bt_psize; 380 root->flags = P_BLEAF; 381 memset(meta, 0, t->bt_psize); 382 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 383 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 384 return (RET_SUCCESS); 385 } 386 387 static int 388 tmp(void) 389 { 390 sigset_t set, oset; 391 int fd, len; 392 char *envtmp = NULL; 393 char path[MAXPATHLEN]; 394 395 if (issetugid() == 0) 396 envtmp = getenv("TMPDIR"); 397 len = snprintf(path, 398 sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp"); 399 if (len < 0 || len >= (int)sizeof(path)) { 400 errno = ENAMETOOLONG; 401 return(-1); 402 } 403 404 (void)sigfillset(&set); 405 (void)__libc_sigprocmask(SIG_BLOCK, &set, &oset); 406 if ((fd = mkostemp(path, O_CLOEXEC)) != -1) 407 (void)unlink(path); 408 (void)__libc_sigprocmask(SIG_SETMASK, &oset, NULL); 409 return(fd); 410 } 411 412 static int 413 byteorder(void) 414 { 415 u_int32_t x; 416 u_char *p; 417 418 x = 0x01020304; 419 p = (u_char *)&x; 420 switch (*p) { 421 case 1: 422 return (BIG_ENDIAN); 423 case 4: 424 return (LITTLE_ENDIAN); 425 default: 426 return (0); 427 } 428 } 429 430 int 431 __bt_fd(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