1/* $NetBSD: bt_open.c,v 1.29 2016/09/24 21:31:25 christos Exp $ */ 2 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 HAVE_NBTOOL_CONFIG_H 36#include "nbtool_config.h" 37#endif 38 39#include <sys/cdefs.h> 40__RCSID("$NetBSD: bt_open.c,v 1.29 2016/09/24 21:31:25 christos Exp $"); 41 42/* 43 * Implementation of btree access method for 4.4BSD. 44 * 45 * The design here was originally based on that of the btree access method 46 * used in the Postgres database system at UC Berkeley. This implementation 47 * is wholly independent of the Postgres code. 48 */ 49 50#include "namespace.h" 51#include <sys/stat.h> 52 53#include <assert.h> 54#include <errno.h> 55#include <fcntl.h> 56#include <limits.h> 57#include <signal.h> 58#include <stdio.h> 59#include <stdlib.h> 60#include <string.h> 61#include <unistd.h> 62#include <paths.h> 63 64#include <db.h> 65#include "btree.h" 66 67#ifdef DEBUG 68#undef MINPSIZE 69#define MINPSIZE 128 70#endif 71 72static int byteorder(void); 73static int nroot(BTREE *); 74 75/* 76 * __BT_OPEN -- Open a btree. 77 * 78 * Creates and fills a DB struct, and calls the routine that actually 79 * opens the btree. 80 * 81 * Parameters: 82 * fname: filename (NULL for in-memory trees) 83 * flags: open flag bits 84 * mode: open permission bits 85 * b: BTREEINFO pointer 86 * 87 * Returns: 88 * NULL on failure, pointer to DB on success. 89 * 90 */ 91DB * 92__bt_open(const char *fname, int flags, mode_t mode, const BTREEINFO *openinfo, 93 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 size_t temp; 103 int machine_lorder; 104 105 t = NULL; 106 107 /* 108 * Intention is to make sure all of the user's selections are okay 109 * here and then use them without checking. Can't be complete, since 110 * we don't know the right page size, lorder or flags until the backing 111 * file is opened. Also, the file's page size can cause the cachesize 112 * to change. 113 */ 114 machine_lorder = byteorder(); 115 if (openinfo) { 116 b = *openinfo; 117 118 /* Flags: R_DUP. */ 119 if (b.flags & ~(R_DUP)) 120 goto einval; 121 122 /* 123 * Page size must be indx_t aligned and >= MINPSIZE. Default 124 * page size is set farther on, based on the underlying file 125 * transfer size. 126 */ 127 if (b.psize && 128 (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 || 129 b.psize & (sizeof(indx_t) - 1))) 130 goto einval; 131 132 /* Minimum number of keys per page; absolute minimum is 2. */ 133 if (b.minkeypage) { 134 if (b.minkeypage < 2) 135 goto einval; 136 } else 137 b.minkeypage = DEFMINKEYPAGE; 138 139 /* If no comparison, use default comparison and prefix. */ 140 if (b.compare == NULL) { 141 b.compare = __bt_defcmp; 142 if (b.prefix == NULL) 143 b.prefix = __bt_defpfx; 144 } 145 146 if (b.lorder == 0) 147 b.lorder = machine_lorder; 148 } else { 149 b.compare = __bt_defcmp; 150 b.cachesize = 0; 151 b.flags = 0; 152 b.lorder = machine_lorder; 153 b.minkeypage = DEFMINKEYPAGE; 154 b.prefix = __bt_defpfx; 155 b.psize = 0; 156 } 157 158 /* Check for the ubiquitous PDP-11. */ 159 if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN) 160 goto einval; 161 162 /* Allocate and initialize DB and BTREE structures. */ 163 if ((t = malloc(sizeof(*t))) == NULL) 164 goto err; 165 memset(t, 0, sizeof(BTREE)); 166 t->bt_fd = -1; /* Don't close unopened fd on error. */ 167 t->bt_lorder = b.lorder; 168 t->bt_order = NOT; 169 t->bt_cmp = b.compare; 170 t->bt_pfx = b.prefix; 171 t->bt_rfd = -1; 172 173 if ((t->bt_dbp = dbp = malloc(sizeof(*dbp))) == NULL) 174 goto err; 175 memset(t->bt_dbp, 0, sizeof(DB)); 176 if (t->bt_lorder != machine_lorder) 177 F_SET(t, B_NEEDSWAP); 178 179 dbp->type = DB_BTREE; 180 dbp->internal = t; 181 dbp->close = __bt_close; 182 dbp->del = __bt_delete; 183 dbp->fd = __bt_fd; 184 dbp->get = __bt_get; 185 dbp->put = __bt_put; 186 dbp->seq = __bt_seq; 187 dbp->sync = __bt_sync; 188 189 /* 190 * If no file name was supplied, this is an in-memory btree and we 191 * open a backing temporary file. Otherwise, it's a disk-based tree. 192 */ 193 if (fname) { 194 switch (flags & O_ACCMODE) { 195 case O_RDONLY: 196 F_SET(t, B_RDONLY); 197 break; 198 case O_RDWR: 199 break; 200 case O_WRONLY: 201 default: 202 goto einval; 203 } 204 if ((t->bt_fd = __dbopen(fname, flags, mode, &sb)) == -1) 205 goto err; 206 } else { 207 if ((flags & O_ACCMODE) != O_RDWR) 208 goto einval; 209 if ((t->bt_fd = __dbtemp("bt.", &sb)) == -1) 210 goto err; 211 F_SET(t, B_INMEM); 212 } 213 214 215 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 = (b.cachesize + t->bt_psize - 1) / 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 temp = (t->bt_psize - BTDATAOFF) / b.minkeypage - 296 (sizeof(indx_t) + NBLEAFDBT(0, 0)); 297 _DBFIT(temp, indx_t); 298 t->bt_ovflsize = (indx_t)temp; 299 if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t)) { 300 size_t l = NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t); 301 _DBFIT(l, indx_t); 302 t->bt_ovflsize = (indx_t)l; 303 } 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 326einval: errno = EINVAL; 327 goto err; 328 329eftype: errno = EFTYPE; 330 goto err; 331 332err: 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 */ 351static int 352nroot(BTREE *t) 353{ 354 PAGE *meta, *root; 355 pgno_t npg; 356 357 if ((root = mpool_get(t->bt_mp, 1, 0)) != NULL) { 358 if (root->lower == 0 && 359 root->pgno == 0 && 360 root->linp[0] == 0) { 361 mpool_delete(t->bt_mp, root); 362 errno = EINVAL; 363 } else { 364 mpool_put(t->bt_mp, root, 0); 365 return RET_SUCCESS; 366 } 367 } 368 if (errno != EINVAL) /* It's OK to not exist. */ 369 return (RET_ERROR); 370 errno = 0; 371 372 if ((meta = mpool_newf(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL) 373 return (RET_ERROR); 374 375 if ((root = mpool_newf(t->bt_mp, &npg, MPOOL_PAGE_NEXT)) == NULL) 376 return (RET_ERROR); 377 378 if (npg != P_ROOT) 379 return (RET_ERROR); 380 root->pgno = npg; 381 root->prevpg = root->nextpg = P_INVALID; 382 root->lower = BTDATAOFF; 383 root->upper = t->bt_psize; 384 root->flags = P_BLEAF; 385 memset(meta, 0, t->bt_psize); 386 mpool_put(t->bt_mp, meta, MPOOL_DIRTY); 387 mpool_put(t->bt_mp, root, MPOOL_DIRTY); 388 return (RET_SUCCESS); 389} 390 391static int 392byteorder(void) 393{ 394 uint32_t x; 395 uint8_t *p; 396 397 x = 0x01020304; 398 p = (uint8_t *)(void *)&x; 399 switch (*p) { 400 case 1: 401 return (BIG_ENDIAN); 402 case 4: 403 return (LITTLE_ENDIAN); 404 default: 405 return (0); 406 } 407} 408 409int 410__bt_fd(const DB *dbp) 411{ 412 BTREE *t; 413 414 t = dbp->internal; 415 416 /* Toss any page pinned across calls. */ 417 if (t->bt_pinned != NULL) { 418 mpool_put(t->bt_mp, t->bt_pinned, 0); 419 t->bt_pinned = NULL; 420 } 421 422 /* In-memory database can't have a file descriptor. */ 423 if (F_ISSET(t, B_INMEM)) { 424 errno = ENOENT; 425 return (-1); 426 } 427 return (t->bt_fd); 428} 429