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 /* 36 * Implementation of btree access method for 4.4BSD. 37 * 38 * The design here was originally based on that of the btree access method 39 * used in the Postgres database system at UC Berkeley. This implementation 40 * is wholly independent of the Postgres code. 41 */ 42 43 #include "namespace.h" 44 #include <sys/param.h> 45 #include <sys/stat.h> 46 47 #include <errno.h> 48 #include <fcntl.h> 49 #include <limits.h> 50 #include <signal.h> 51 #include <stdio.h> 52 #include <stdlib.h> 53 #include <string.h> 54 #include <unistd.h> 55 #include "un-namespace.h" 56 #include "libc_private.h" 57 58 #include <db.h> 59 #include "btree.h" 60 61 #ifdef DEBUG 62 #undef MINPSIZE 63 #define MINPSIZE 128 64 #endif 65 66 static int byteorder(void); 67 static int nroot(BTREE *); 68 static int tmp(void); 69 70 /* 71 * __BT_OPEN -- Open a btree. 72 * 73 * Creates and fills a DB struct, and calls the routine that actually 74 * opens the btree. 75 * 76 * Parameters: 77 * fname: filename (NULL for in-memory trees) 78 * flags: open flag bits 79 * mode: open permission bits 80 * b: BTREEINFO pointer 81 * 82 * Returns: 83 * NULL on failure, pointer to DB on success. 84 * 85 */ 86 DB * 87 __bt_open(const char *fname, int flags, int mode, const BTREEINFO *openinfo, int dflags) 88 { 89 struct stat sb; 90 BTMETA m; 91 BTREE *t; 92 BTREEINFO b; 93 DB *dbp; 94 pgno_t ncache; 95 ssize_t nr; 96 int machine_lorder, saved_errno; 97 98 t = NULL; 99 100 /* 101 * Intention is to make sure all of the user's selections are okay 102 * here and then use them without checking. Can't be complete, since 103 * we don't know the right page size, lorder or flags until the backing 104 * file is opened. Also, the file's page size can cause the cachesize 105 * to change. 106 */ 107 machine_lorder = byteorder(); 108 if (openinfo) { 109 b = *openinfo; 110 111 /* Flags: R_DUP. */ 112 if (b.flags & ~(R_DUP)) 113 goto einval; 114 115 /* 116 * Page size must be indx_t aligned and >= MINPSIZE. Default 117 * page size is set farther on, based on the underlying file 118 * transfer size. 119 */ 120 if (b.psize && 121 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 122 b.psize & (sizeof(indx_t) - 1) )) 123 goto einval; 124 125 /* Minimum number of keys per page; absolute minimum is 2. */ 126 if (b.minkeypage) { 127 if (b.minkeypage < 2) 128 goto einval; 129 } else 130 b.minkeypage = DEFMINKEYPAGE; 131 132 /* If no comparison, use default comparison and prefix. */ 133 if (b.compare == NULL) { 134 b.compare = __bt_defcmp; 135 if (b.prefix == NULL) 136 b.prefix = __bt_defpfx; 137 } 138 139 if (b.lorder == 0) 140 b.lorder = machine_lorder; 141 } else { 142 b.compare = __bt_defcmp; 143 b.cachesize = 0; 144 b.flags = 0; 145 b.lorder = machine_lorder; 146 b.minkeypage = DEFMINKEYPAGE; 147 b.prefix = __bt_defpfx; 148 b.psize = 0; 149 } 150 151 /* Check for the ubiquitous PDP-11. */ 152 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 153 goto einval; 154 155 /* Allocate and initialize DB and BTREE structures. */ 156 if ((t = (BTREE *)calloc(1, sizeof(BTREE))) == NULL) 157 goto err; 158 t->bt_fd = -1; /* Don't close unopened fd on error. */ 159 t->bt_lorder = b.lorder; 160 t->bt_order = NOT; 161 t->bt_cmp = b.compare; 162 t->bt_pfx = b.prefix; 163 t->bt_rfd = -1; 164 165 if ((t->bt_dbp = dbp = (DB *)calloc(1, sizeof(DB))) == NULL) 166 goto err; 167 if (t->bt_lorder != machine_lorder) 168 F_SET(t, B_NEEDSWAP); 169 170 dbp->type = DB_BTREE; 171 dbp->internal = t; 172 dbp->close = __bt_close; 173 dbp->del = __bt_delete; 174 dbp->fd = __bt_fd; 175 dbp->get = __bt_get; 176 dbp->put = __bt_put; 177 dbp->seq = __bt_seq; 178 dbp->sync = __bt_sync; 179 180 /* 181 * If no file name was supplied, this is an in-memory btree and we 182 * open a backing temporary file. Otherwise, it's a disk-based tree. 183 */ 184 if (fname) { 185 switch (flags & O_ACCMODE) { 186 case O_RDONLY: 187 F_SET(t, B_RDONLY); 188 break; 189 case O_RDWR: 190 break; 191 case O_WRONLY: 192 default: 193 goto einval; 194 } 195 196 if ((t->bt_fd = _open(fname, flags | O_CLOEXEC, mode)) < 0) 197 goto err; 198 199 } else { 200 if ((flags & O_ACCMODE) != O_RDWR) 201 goto einval; 202 if ((t->bt_fd = tmp()) == -1) 203 goto err; 204 F_SET(t, B_INMEM); 205 } 206 207 if (_fstat(t->bt_fd, &sb)) 208 goto err; 209 if (sb.st_size) { 210 if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0) 211 goto err; 212 if (nr != sizeof(BTMETA)) 213 goto eftype; 214 215 /* 216 * Read in the meta-data. This can change the notion of what 217 * the lorder, page size and flags are, and, when the page size 218 * changes, the cachesize value can change too. If the user 219 * specified the wrong byte order for an existing database, we 220 * don't bother to return an error, we just clear the NEEDSWAP 221 * bit. 222 */ 223 if (m.magic == BTREEMAGIC) 224 F_CLR(t, B_NEEDSWAP); 225 else { 226 F_SET(t, B_NEEDSWAP); 227 M_32_SWAP(m.magic); 228 M_32_SWAP(m.version); 229 M_32_SWAP(m.psize); 230 M_32_SWAP(m.free); 231 M_32_SWAP(m.nrecs); 232 M_32_SWAP(m.flags); 233 } 234 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION) 235 goto eftype; 236 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 || 237 m.psize & (sizeof(indx_t) - 1) ) 238 goto eftype; 239 if (m.flags & ~SAVEMETA) 240 goto eftype; 241 b.psize = m.psize; 242 F_SET(t, m.flags); 243 t->bt_free = m.free; 244 t->bt_nrecs = m.nrecs; 245 } else { 246 /* 247 * Set the page size to the best value for I/O to this file. 248 * Don't overflow the page offset type. 249 */ 250 if (b.psize == 0) { 251 b.psize = sb.st_blksize; 252 if (b.psize < MINPSIZE) 253 b.psize = MINPSIZE; 254 if (b.psize > MAX_PAGE_OFFSET + 1) 255 b.psize = MAX_PAGE_OFFSET + 1; 256 } 257 258 /* Set flag if duplicates permitted. */ 259 if (!(b.flags & R_DUP)) 260 F_SET(t, B_NODUPS); 261 262 t->bt_free = P_INVALID; 263 t->bt_nrecs = 0; 264 F_SET(t, B_METADIRTY); 265 } 266 267 t->bt_psize = b.psize; 268 269 /* Set the cache size; must be a multiple of the page size. */ 270 if (b.cachesize && b.cachesize & (b.psize - 1) ) 271 b.cachesize += (~b.cachesize & (b.psize - 1) ) + 1; 272 if (b.cachesize < b.psize * MINCACHE) 273 b.cachesize = b.psize * MINCACHE; 274 275 /* Calculate number of pages to cache. */ 276 ncache = howmany(b.cachesize, t->bt_psize); 277 278 /* 279 * The btree data structure requires that at least two keys can fit on 280 * a page, but other than that there's no fixed requirement. The user 281 * specified a minimum number per page, and we translated that into the 282 * number of bytes a key/data pair can use before being placed on an 283 * overflow page. This calculation includes the page header, the size 284 * of the index referencing the leaf item and the size of the leaf item 285 * structure. Also, don't let the user specify a minkeypage such that 286 * a key/data pair won't fit even if both key and data are on overflow 287 * pages. 288 */ 289 t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage - 290 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 291 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) 292 t->bt_ovflsize = 293 NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 294 295 /* Initialize the buffer pool. */ 296 if ((t->bt_mp = 297 mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL) 298 goto err; 299 if (!F_ISSET(t, B_INMEM)) 300 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t); 301 302 /* Create a root page if new tree. */ 303 if (nroot(t) == RET_ERROR) 304 goto err; 305 306 /* Global flags. */ 307 if (dflags & DB_LOCK) 308 F_SET(t, B_DB_LOCK); 309 if (dflags & DB_SHMEM) 310 F_SET(t, B_DB_SHMEM); 311 if (dflags & DB_TXN) 312 F_SET(t, B_DB_TXN); 313 314 return (dbp); 315 316 einval: errno = EINVAL; 317 goto err; 318 319 eftype: errno = EFTYPE; 320 goto err; 321 322 err: saved_errno = errno; 323 if (t) { 324 if (t->bt_dbp) 325 free(t->bt_dbp); 326 if (t->bt_fd != -1) 327 (void)_close(t->bt_fd); 328 free(t); 329 } 330 errno = saved_errno; 331 return (NULL); 332 } 333 334 /* 335 * NROOT -- Create the root of a new tree. 336 * 337 * Parameters: 338 * t: tree 339 * 340 * Returns: 341 * RET_ERROR, RET_SUCCESS 342 */ 343 static int 344 nroot(BTREE *t) 345 { 346 PAGE *meta, *root; 347 pgno_t npg; 348 349 if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) { 350 if (root->lower == 0 && 351 root->pgno == 0 && 352 root->linp[0] == 0) { 353 mpool_delete(t->bt_mp, root); 354 errno = EINVAL; 355 } else { 356 mpool_put(t->bt_mp, root, 0); 357 return (RET_SUCCESS); 358 } 359 } 360 if (errno != EINVAL) /* It's OK to not exist. */ 361 return (RET_ERROR); 362 errno = 0; 363 364 if ((meta = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL) 365 return (RET_ERROR); 366 367 if ((root = mpool_new(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL) 368 return (RET_ERROR); 369 370 if (npg != P_ROOT) 371 return (RET_ERROR); 372 root->pgno = npg; 373 root->prevpg = root->nextpg = P_INVALID; 374 root->lower = BTDATAOFF; 375 root->upper = t->bt_psize; 376 root->flags = P_BLEAF; 377 memset(meta, 0, t->bt_psize); 378 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 379 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 380 return (RET_SUCCESS); 381 } 382 383 static int 384 tmp(void) 385 { 386 sigset_t set, oset; 387 int fd, len; 388 char *envtmp; 389 char path[MAXPATHLEN]; 390 391 envtmp = secure_getenv("TMPDIR"); 392 len = snprintf(path, 393 sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp"); 394 if (len < 0 || len >= (int)sizeof(path)) { 395 errno = ENAMETOOLONG; 396 return(-1); 397 } 398 399 (void)sigfillset(&set); 400 (void)__libc_sigprocmask(SIG_BLOCK, &set, &oset); 401 if ((fd = mkostemp(path, O_CLOEXEC)) != -1) 402 (void)unlink(path); 403 (void)__libc_sigprocmask(SIG_SETMASK, &oset, NULL); 404 return(fd); 405 } 406 407 static int 408 byteorder(void) 409 { 410 u_int32_t x; 411 u_char *p; 412 413 x = 0x01020304; 414 p = (u_char *)&x; 415 switch (*p) { 416 case 1: 417 return (BIG_ENDIAN); 418 case 4: 419 return (LITTLE_ENDIAN); 420 default: 421 return (0); 422 } 423 } 424 425 int 426 __bt_fd(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