1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1990, 1993, 1994 5 * The Regents of the University of California. All rights reserved. 6 * 7 * This code is derived from software contributed to Berkeley by 8 * Mike Olson. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 */ 34 35 #if defined(LIBC_SCCS) && !defined(lint) 36 static char sccsid[] = "@(#)bt_open.c 8.10 (Berkeley) 8/17/94"; 37 #endif /* LIBC_SCCS and not lint */ 38 #include <sys/cdefs.h> 39 __FBSDID("$FreeBSD$"); 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 "namespace.h" 50 #include <sys/param.h> 51 #include <sys/stat.h> 52 53 #include <errno.h> 54 #include <fcntl.h> 55 #include <limits.h> 56 #include <signal.h> 57 #include <stdio.h> 58 #include <stdlib.h> 59 #include <string.h> 60 #include <unistd.h> 61 #include "un-namespace.h" 62 #include "libc_private.h" 63 64 #include <db.h> 65 #include "btree.h" 66 67 #ifdef DEBUG 68 #undef MINPSIZE 69 #define MINPSIZE 128 70 #endif 71 72 static int byteorder(void); 73 static int nroot(BTREE *); 74 static int tmp(void); 75 76 /* 77 * __BT_OPEN -- Open a btree. 78 * 79 * Creates and fills a DB struct, and calls the routine that actually 80 * opens the btree. 81 * 82 * Parameters: 83 * fname: filename (NULL for in-memory trees) 84 * flags: open flag bits 85 * mode: open permission bits 86 * b: BTREEINFO pointer 87 * 88 * Returns: 89 * NULL on failure, pointer to DB on success. 90 * 91 */ 92 DB * 93 __bt_open(const char *fname, int flags, int mode, const BTREEINFO *openinfo, int dflags) 94 { 95 struct stat sb; 96 BTMETA m; 97 BTREE *t; 98 BTREEINFO b; 99 DB *dbp; 100 pgno_t ncache; 101 ssize_t nr; 102 int machine_lorder, saved_errno; 103 104 t = NULL; 105 106 /* 107 * Intention is to make sure all of the user's selections are okay 108 * here and then use them without checking. Can't be complete, since 109 * we don't know the right page size, lorder or flags until the backing 110 * file is opened. Also, the file's page size can cause the cachesize 111 * to change. 112 */ 113 machine_lorder = byteorder(); 114 if (openinfo) { 115 b = *openinfo; 116 117 /* Flags: R_DUP. */ 118 if (b.flags & ~(R_DUP)) 119 goto einval; 120 121 /* 122 * Page size must be indx_t aligned and >= MINPSIZE. Default 123 * page size is set farther on, based on the underlying file 124 * transfer size. 125 */ 126 if (b.psize && 127 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 128 b.psize & (sizeof(indx_t) - 1) )) 129 goto einval; 130 131 /* Minimum number of keys per page; absolute minimum is 2. */ 132 if (b.minkeypage) { 133 if (b.minkeypage < 2) 134 goto einval; 135 } else 136 b.minkeypage = DEFMINKEYPAGE; 137 138 /* If no comparison, use default comparison and prefix. */ 139 if (b.compare == NULL) { 140 b.compare = __bt_defcmp; 141 if (b.prefix == NULL) 142 b.prefix = __bt_defpfx; 143 } 144 145 if (b.lorder == 0) 146 b.lorder = machine_lorder; 147 } else { 148 b.compare = __bt_defcmp; 149 b.cachesize = 0; 150 b.flags = 0; 151 b.lorder = machine_lorder; 152 b.minkeypage = DEFMINKEYPAGE; 153 b.prefix = __bt_defpfx; 154 b.psize = 0; 155 } 156 157 /* Check for the ubiquitous PDP-11. */ 158 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 159 goto einval; 160 161 /* Allocate and initialize DB and BTREE structures. */ 162 if ((t = (BTREE *)calloc(1, sizeof(BTREE))) == NULL) 163 goto err; 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 *)calloc(1, sizeof(DB))) == NULL) 172 goto err; 173 if (t->bt_lorder != machine_lorder) 174 F_SET(t, B_NEEDSWAP); 175 176 dbp->type = DB_BTREE; 177 dbp->internal = t; 178 dbp->close = __bt_close; 179 dbp->del = __bt_delete; 180 dbp->fd = __bt_fd; 181 dbp->get = __bt_get; 182 dbp->put = __bt_put; 183 dbp->seq = __bt_seq; 184 dbp->sync = __bt_sync; 185 186 /* 187 * If no file name was supplied, this is an in-memory btree and we 188 * open a backing temporary file. Otherwise, it's a disk-based tree. 189 */ 190 if (fname) { 191 switch (flags & O_ACCMODE) { 192 case O_RDONLY: 193 F_SET(t, B_RDONLY); 194 break; 195 case O_RDWR: 196 break; 197 case O_WRONLY: 198 default: 199 goto einval; 200 } 201 202 if ((t->bt_fd = _open(fname, flags | O_CLOEXEC, mode)) < 0) 203 goto err; 204 205 } else { 206 if ((flags & O_ACCMODE) != O_RDWR) 207 goto einval; 208 if ((t->bt_fd = tmp()) == -1) 209 goto err; 210 F_SET(t, B_INMEM); 211 } 212 213 if (_fstat(t->bt_fd, &sb)) 214 goto err; 215 if (sb.st_size) { 216 if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 217 goto err; 218 if (nr != sizeof(BTMETA)) 219 goto eftype; 220 221 /* 222 * Read in the meta-data. This can change the notion of what 223 * the lorder, page size and flags are, and, when the page size 224 * changes, the cachesize value can change too. If the user 225 * specified the wrong byte order for an existing database, we 226 * don't bother to return an error, we just clear the NEEDSWAP 227 * bit. 228 */ 229 if (m.magic == BTREEMAGIC) 230 F_CLR(t, B_NEEDSWAP); 231 else { 232 F_SET(t, B_NEEDSWAP); 233 M_32_SWAP(m.magic); 234 M_32_SWAP(m.version); 235 M_32_SWAP(m.psize); 236 M_32_SWAP(m.free); 237 M_32_SWAP(m.nrecs); 238 M_32_SWAP(m.flags); 239 } 240 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 241 goto eftype; 242 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 243 m.psize & (sizeof(indx_t) - 1) ) 244 goto eftype; 245 if (m.flags & ~SAVEMETA) 246 goto eftype; 247 b.psize = m.psize; 248 F_SET(t, m.flags); 249 t->bt_free = m.free; 250 t->bt_nrecs = m.nrecs; 251 } else { 252 /* 253 * Set the page size to the best value for I/O to this file. 254 * Don't overflow the page offset type. 255 */ 256 if (b.psize == 0) { 257 b.psize = sb.st_blksize; 258 if (b.psize < MINPSIZE) 259 b.psize = MINPSIZE; 260 if (b.psize > MAX_PAGE_OFFSET + 1) 261 b.psize = MAX_PAGE_OFFSET + 1; 262 } 263 264 /* Set flag if duplicates permitted. */ 265 if (!(b.flags & R_DUP)) 266 F_SET(t, B_NODUPS); 267 268 t->bt_free = P_INVALID; 269 t->bt_nrecs = 0; 270 F_SET(t, B_METADIRTY); 271 } 272 273 t->bt_psize = b.psize; 274 275 /* Set the cache size; must be a multiple of the page size. */ 276 if (b.cachesize && b.cachesize & (b.psize - 1) ) 277 b.cachesize += (~b.cachesize & (b.psize - 1) ) + 1; 278 if (b.cachesize < b.psize * MINCACHE) 279 b.cachesize = b.psize * MINCACHE; 280 281 /* Calculate number of pages to cache. */ 282 ncache = howmany(b.cachesize, t->bt_psize); 283 284 /* 285 * The btree data structure requires that at least two keys can fit on 286 * a page, but other than that there's no fixed requirement. The user 287 * specified a minimum number per page, and we translated that into the 288 * number of bytes a key/data pair can use before being placed on an 289 * overflow page. This calculation includes the page header, the size 290 * of the index referencing the leaf item and the size of the leaf item 291 * structure. Also, don't let the user specify a minkeypage such that 292 * a key/data pair won't fit even if both key and data are on overflow 293 * pages. 294 */ 295 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 296 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 297 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 298 t->bt_ovflsize = 299 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 300 301 /* Initialize the buffer pool. */ 302 if ((t->bt_mp = 303 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 304 goto err; 305 if (!F_ISSET(t, B_INMEM)) 306 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 307 308 /* Create a root page if new tree. */ 309 if (nroot(t) == RET_ERROR) 310 goto err; 311 312 /* Global flags. */ 313 if (dflags & DB_LOCK) 314 F_SET(t, B_DB_LOCK); 315 if (dflags & DB_SHMEM) 316 F_SET(t, B_DB_SHMEM); 317 if (dflags & DB_TXN) 318 F_SET(t, B_DB_TXN); 319 320 return (dbp); 321 322 einval: errno = EINVAL; 323 goto err; 324 325 eftype: errno = EFTYPE; 326 goto err; 327 328 err: saved_errno = errno; 329 if (t) { 330 if (t->bt_dbp) 331 free(t->bt_dbp); 332 if (t->bt_fd != -1) 333 (void)_close(t->bt_fd); 334 free(t); 335 } 336 errno = saved_errno; 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 */ 349 static int 350 nroot(BTREE *t) 351 { 352 PAGE *meta, *root; 353 pgno_t npg; 354 355 if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) { 356 if (root->lower == 0 && 357 root->pgno == 0 && 358 root->linp[0] == 0) { 359 mpool_delete(t->bt_mp, root); 360 errno = EINVAL; 361 } else { 362 mpool_put(t->bt_mp, root, 0); 363 return (RET_SUCCESS); 364 } 365 } 366 if (errno != EINVAL) /* It's OK to not exist. */ 367 return (RET_ERROR); 368 errno = 0; 369 370 if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL) 371 return (RET_ERROR); 372 373 if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL) 374 return (RET_ERROR); 375 376 if (npg != P_ROOT) 377 return (RET_ERROR); 378 root->pgno = npg; 379 root->prevpg = root->nextpg = P_INVALID; 380 root->lower = BTDATAOFF; 381 root->upper = t->bt_psize; 382 root->flags = P_BLEAF; 383 memset(meta, 0, t->bt_psize); 384 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 385 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 386 return (RET_SUCCESS); 387 } 388 389 static int 390 tmp(void) 391 { 392 sigset_t set, oset; 393 int fd, len; 394 char *envtmp = NULL; 395 char path[MAXPATHLEN]; 396 397 if (issetugid() == 0) 398 envtmp = getenv("TMPDIR"); 399 len = snprintf(path, 400 sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp"); 401 if (len < 0 || len >= (int)sizeof(path)) { 402 errno = ENAMETOOLONG; 403 return(-1); 404 } 405 406 (void)sigfillset(&set); 407 (void)__libc_sigprocmask(SIG_BLOCK, &set, &oset); 408 if ((fd = mkostemp(path, O_CLOEXEC)) != -1) 409 (void)unlink(path); 410 (void)__libc_sigprocmask(SIG_SETMASK, &oset, NULL); 411 return(fd); 412 } 413 414 static int 415 byteorder(void) 416 { 417 u_int32_t x; 418 u_char *p; 419 420 x = 0x01020304; 421 p = (u_char *)&x; 422 switch (*p) { 423 case 1: 424 return (BIG_ENDIAN); 425 case 4: 426 return (LITTLE_ENDIAN); 427 default: 428 return (0); 429 } 430 } 431 432 int 433 __bt_fd(const DB *dbp) 434 { 435 BTREE *t; 436 437 t = dbp->internal; 438 439 /* Toss any page pinned across calls. */ 440 if (t->bt_pinned != NULL) { 441 mpool_put(t->bt_mp, t->bt_pinned, 0); 442 t->bt_pinned = NULL; 443 } 444 445 /* In-memory database can't have a file descriptor. */ 446 if (F_ISSET(t, B_INMEM)) { 447 errno = ENOENT; 448 return (-1); 449 } 450 return (t->bt_fd); 451 } 452