db/btree/bt_open.c

   1 /*
   2  * Copyright (c) 1999 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * Copyright (c) 1999-2003 Apple Computer, Inc.  All Rights Reserved.
   7  *
   8  * This file contains Original Code and/or Modifications of Original Code
   9  * as defined in and that are subject to the Apple Public Source License
  10  * Version 2.0 (the 'License'). You may not use this file except in
  11  * compliance with the License. Please obtain a copy of the License at
  12  * http://www.opensource.apple.com/apsl/ and read it before using this
  13  * file.
  14  *
  15  * The Original Code and all software distributed under the License are
  16  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  17  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  18  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  20  * Please see the License for the specific language governing rights and
  21  * limitations under the License.
  22  *
  23  * @APPLE_LICENSE_HEADER_END@
  24  */
  25 /*
  26  * Copyright (c) 1990, 1993
  27  *      The Regents of the University of California.  All rights reserved.
  28  *
  29  * This code is derived from software contributed to Berkeley by
  30  * Mike Olson.
  31  *
  32  * Redistribution and use in source and binary forms, with or without
  33  * modification, are permitted provided that the following conditions
  34  * are met:
  35  * 1. Redistributions of source code must retain the above copyright
  36  *    notice, this list of conditions and the following disclaimer.
  37  * 2. Redistributions in binary form must reproduce the above copyright
  38  *    notice, this list of conditions and the following disclaimer in the
  39  *    documentation and/or other materials provided with the distribution.
  40  * 3. All advertising materials mentioning features or use of this software
  41  *    must display the following acknowledgement:
  42  *      This product includes software developed by the University of
  43  *      California, Berkeley and its contributors.
  44  * 4. Neither the name of the University nor the names of its contributors
  45  *    may be used to endorse or promote products derived from this software
  46  *    without specific prior written permission.
  47  *
  48  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  49  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  50  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  51  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  52  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  53  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  54  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  55  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  56  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  57  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  58  * SUCH DAMAGE.
  59  */
  60
  61
  62 /*
  63  * Implementation of btree access method for 4.4BSD.
  64  *
  65  * The design here was originally based on that of the btree access method
  66  * used in the Postgres database system at UC Berkeley.  This implementation
  67  * is wholly independent of the Postgres code.
  68  */
  69
  70 #include <sys/param.h>
  71 #include <sys/stat.h>
  72
  73 #include <errno.h>
  74 #include <fcntl.h>
  75 #include <limits.h>
  76 #include <signal.h>
  77 #include <stdio.h>
  78 #include <stdlib.h>
  79 #include <string.h>
  80 #include <unistd.h>
  81
  82 #include <db.h>
  83 #include "btree.h"
  84
  85 static int byteorder __P((void));
  86 static int nroot __P((BTREE *));
  87 static int tmp __P((void));
  88
  89 /*
  90  * __BT_OPEN -- Open a btree.
  91  *
  92  * Creates and fills a DB struct, and calls the routine that actually
  93  * opens the btree.
  94  *
  95  * Parameters:
  96  *      fname:  filename (NULL for in-memory trees)
  97  *      flags:  open flag bits
  98  *      mode:   open permission bits
  99  *      b:      BTREEINFO pointer
 100  *
 101  * Returns:
 102  *      NULL on failure, pointer to DB on success.
 103  *
 104  */
 105 DB *
 106 __bt_open(fname, flags, mode, openinfo, dflags)
 107         const char *fname;
 108         int flags, mode, dflags;
 109         const BTREEINFO *openinfo;
 110 {
 111         struct stat sb;
 112         BTMETA m;
 113         BTREE *t;
 114         BTREEINFO b;
 115         DB *dbp;
 116         pgno_t ncache;
 117         ssize_t nr;
 118         int machine_lorder;
 119
 120         t = NULL;
 121
 122         /*
 123          * Intention is to make sure all of the user's selections are okay
 124          * here and then use them without checking.  Can't be complete, since
 125          * we don't know the right page size, lorder or flags until the backing
 126          * file is opened.  Also, the file's page size can cause the cachesize
 127          * to change.
 128          */
 129         machine_lorder = byteorder();
 130         if (openinfo) {
 131                 b = *openinfo;
 132
 133                 /* Flags: R_DUP. */
 134                 if (b.flags & ~(R_DUP))
 135                         goto einval;
 136
 137                 /*
 138                  * Page size must be indx_t aligned and >= MINPSIZE.  Default
 139                  * page size is set farther on, based on the underlying file
 140                  * transfer size.
 141                  */
 142                 if (b.psize &&
 143                     (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
 144                     b.psize & sizeof(indx_t) - 1))
 145                         goto einval;
 146
 147                 /* Minimum number of keys per page; absolute minimum is 2. */
 148                 if (b.minkeypage) {
 149                         if (b.minkeypage < 2)
 150                                 goto einval;
 151                 } else
 152                         b.minkeypage = DEFMINKEYPAGE;
 153
 154                 /* If no comparison, use default comparison and prefix. */
 155                 if (b.compare == NULL) {
 156                         b.compare = __bt_defcmp;
 157                         if (b.prefix == NULL)
 158                                 b.prefix = __bt_defpfx;
 159                 }
 160
 161                 if (b.lorder == 0)
 162                         b.lorder = machine_lorder;
 163         } else {
 164                 b.compare = __bt_defcmp;
 165                 b.cachesize = 0;
 166                 b.flags = 0;
 167                 b.lorder = machine_lorder;
 168                 b.minkeypage = DEFMINKEYPAGE;
 169                 b.prefix = __bt_defpfx;
 170                 b.psize = 0;
 171         }
 172
 173         /* Check for the ubiquitous PDP-11. */
 174         if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN)
 175                 goto einval;
 176
 177         /* Allocate and initialize DB and BTREE structures. */
 178         if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL)
 179                 goto err;
 180         memset(t, 0, sizeof(BTREE));
 181         t->bt_bcursor.pgno = P_INVALID;
 182         t->bt_fd = -1;                  /* Don't close unopened fd on error. */
 183         t->bt_lorder = b.lorder;
 184         t->bt_order = NOT;
 185         t->bt_cmp = b.compare;
 186         t->bt_pfx = b.prefix;
 187         t->bt_rfd = -1;
 188
 189         if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL)
 190                 goto err;
 191         t->bt_flags = 0;
 192         if (t->bt_lorder != machine_lorder)
 193                 SET(t, B_NEEDSWAP);
 194
 195         dbp->type = DB_BTREE;
 196         dbp->internal = t;
 197         dbp->close = __bt_close;
 198         dbp->del = __bt_delete;
 199         dbp->fd = __bt_fd;
 200         dbp->get = __bt_get;
 201         dbp->put = __bt_put;
 202         dbp->seq = __bt_seq;
 203         dbp->sync = __bt_sync;
 204
 205         /*
 206          * If no file name was supplied, this is an in-memory btree and we
 207          * open a backing temporary file.  Otherwise, it's a disk-based tree.
 208          */
 209         if (fname) {
 210                 switch(flags & O_ACCMODE) {
 211                 case O_RDONLY:
 212                         SET(t, B_RDONLY);
 213                         break;
 214                 case O_RDWR:
 215                         break;
 216                 case O_WRONLY:
 217                 default:
 218                         goto einval;
 219                 }
 220
 221                 if ((t->bt_fd = open(fname, flags, mode)) < 0)
 222                         goto err;
 223
 224         } else {
 225                 if ((flags & O_ACCMODE) != O_RDWR)
 226                         goto einval;
 227                 if ((t->bt_fd = tmp()) == -1)
 228                         goto err;
 229                 SET(t, B_INMEM);
 230         }
 231
 232         if (fcntl(t->bt_fd, F_SETFD, 1) == -1)
 233                 goto err;
 234
 235         if (fstat(t->bt_fd, &sb))
 236                 goto err;
 237         if (sb.st_size) {
 238                 nr = read(t->bt_fd, &m, sizeof(BTMETA));
 239                 if (nr < 0)
 240                         goto err;
 241                 if (nr != sizeof(BTMETA))
 242                         goto eftype;
 243
 244                 /*
 245                  * Read in the meta-data.  This can change the notion of what
 246                  * the lorder, page size and flags are, and, when the page size
 247                  * changes, the cachesize value can change too.  If the user
 248                  * specified the wrong byte order for an existing database, we
 249                  * don't bother to return an error, we just clear the NEEDSWAP
 250                  * bit.
 251                  */
 252                 if (m.m_magic == BTREEMAGIC)
 253                         CLR(t, B_NEEDSWAP);
 254                 else {
 255                         SET(t, B_NEEDSWAP);
 256                         M_32_SWAP(m.m_magic);
 257                         M_32_SWAP(m.m_version);
 258                         M_32_SWAP(m.m_psize);
 259                         M_32_SWAP(m.m_free);
 260                         M_32_SWAP(m.m_nrecs);
 261                         M_32_SWAP(m.m_flags);
 262                 }
 263                 if (m.m_magic != BTREEMAGIC || m.m_version != BTREEVERSION)
 264                         goto eftype;
 265                 if (m.m_psize < MINPSIZE || m.m_psize > MAX_PAGE_OFFSET + 1 ||
 266                     m.m_psize & sizeof(indx_t) - 1)
 267                         goto eftype;
 268                 if (m.m_flags & ~SAVEMETA)
 269                         goto eftype;
 270                 b.psize = m.m_psize;
 271                 t->bt_flags |= m.m_flags;
 272                 t->bt_free = m.m_free;
 273                 t->bt_nrecs = m.m_nrecs;
 274         } else {
 275                 /*
 276                  * Set the page size to the best value for I/O to this file.
 277                  * Don't overflow the page offset type.
 278                  */
 279                 if (b.psize == 0) {
 280                         b.psize = sb.st_blksize;
 281                         if (b.psize < MINPSIZE)
 282                                 b.psize = MINPSIZE;
 283                         if (b.psize > MAX_PAGE_OFFSET + 1)
 284                                 b.psize = MAX_PAGE_OFFSET + 1;
 285                 }
 286
 287                 /* Set flag if duplicates permitted. */
 288                 if (!(b.flags & R_DUP))
 289                         SET(t, B_NODUPS);
 290
 291                 t->bt_free = P_INVALID;
 292                 t->bt_nrecs = 0;
 293                 SET(t, B_METADIRTY);
 294         }
 295
 296         t->bt_psize = b.psize;
 297
 298         /* Set the cache size; must be a multiple of the page size. */
 299         if (b.cachesize && b.cachesize & b.psize - 1)
 300                 b.cachesize += (~b.cachesize & b.psize - 1) + 1;
 301         if (b.cachesize < b.psize * MINCACHE)
 302                 b.cachesize = b.psize * MINCACHE;
 303
 304         /* Calculate number of pages to cache. */
 305         ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
 306
 307         /*
 308          * The btree data structure requires that at least two keys can fit on
 309          * a page, but other than that there's no fixed requirement.  The user
 310          * specified a minimum number per page, and we translated that into the
 311          * number of bytes a key/data pair can use before being placed on an
 312          * overflow page.  This calculation includes the page header, the size
 313          * of the index referencing the leaf item and the size of the leaf item
 314          * structure.  Also, don't let the user specify a minkeypage such that
 315          * a key/data pair won't fit even if both key and data are on overflow
 316          * pages.
 317          */
 318         t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage -
 319             (sizeof(indx_t) + NBLEAFDBT(0, 0));
 320         if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t))
 321                 t->bt_ovflsize =
 322                     NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
 323
 324         /* Initialize the buffer pool. */
 325         if ((t->bt_mp =
 326             mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
 327                 goto err;
 328         if (!ISSET(t, B_INMEM))
 329                 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t);
 330
 331         /* Create a root page if new tree. */
 332         if (nroot(t) == RET_ERROR)
 333                 goto err;
 334
 335         /* Global flags. */
 336         if (dflags & DB_LOCK)
 337                 SET(t, B_DB_LOCK);
 338         if (dflags & DB_SHMEM)
 339                 SET(t, B_DB_SHMEM);
 340         if (dflags & DB_TXN)
 341                 SET(t, B_DB_TXN);
 342
 343         return (dbp);
 344
 345 einval: errno = EINVAL;
 346         goto err;
 347
 348 eftype: errno = EFTYPE;
 349         goto err;
 350
 351 err:    if (t) {
 352                 if (t->bt_dbp)
 353                         free(t->bt_dbp);
 354                 if (t->bt_fd != -1)
 355                         (void)close(t->bt_fd);
 356                 free(t);
 357         }
 358         return (NULL);
 359 }
 360
 361 /*
 362  * NROOT -- Create the root of a new tree.
 363  *
 364  * Parameters:
 365  *      t:      tree
 366  *
 367  * Returns:
 368  *      RET_ERROR, RET_SUCCESS
 369  */
 370 static int
 371 nroot(t)
 372         BTREE *t;
 373 {
 374         PAGE *meta, *root;
 375         pgno_t npg;
 376
 377         if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) {
 378                 mpool_put(t->bt_mp, meta, 0);
 379                 return (RET_SUCCESS);
 380         }
 381         if (errno != EINVAL)
 382                 return (RET_ERROR);
 383
 384         if ((meta = mpool_new(t->bt_mp, &npg)) == NULL)
 385                 return (RET_ERROR);
 386
 387         if ((root = mpool_new(t->bt_mp, &npg)) == NULL)
 388                 return (RET_ERROR);
 389
 390         if (npg != P_ROOT)
 391                 return (RET_ERROR);
 392         root->pgno = npg;
 393         root->prevpg = root->nextpg = P_INVALID;
 394         root->lower = BTDATAOFF;
 395         root->upper = t->bt_psize;
 396         root->flags = P_BLEAF;
 397         memset(meta, 0, t->bt_psize);
 398         mpool_put(t->bt_mp, meta, MPOOL_DIRTY);
 399         mpool_put(t->bt_mp, root, MPOOL_DIRTY);
 400         return (RET_SUCCESS);
 401 }
 402
 403 static int
 404 tmp()
 405 {
 406         sigset_t set, oset;
 407         int fd;
 408         char *envtmp;
 409         char path[MAXPATHLEN];
 410
 411         envtmp = getenv("TMPDIR");
 412         (void)snprintf(path,
 413             sizeof(path), "%s/bt.XXXXXX", envtmp ? envtmp : "/tmp");
 414
 415         (void)sigfillset(&set);
 416         (void)sigprocmask(SIG_BLOCK, &set, &oset);
 417         if ((fd = mkstemp(path)) != -1)
 418                 (void)unlink(path);
 419         (void)sigprocmask(SIG_SETMASK, &oset, NULL);
 420         return(fd);
 421 }
 422
 423 static int
 424 byteorder()
 425 {
 426         u_int32_t x;
 427         u_char *p;
 428
 429         x = 0x01020304;
 430         p = (u_char *)&x;
 431         switch (*p) {
 432         case 1:
 433                 return (BIG_ENDIAN);
 434         case 4:
 435                 return (LITTLE_ENDIAN);
 436         default:
 437                 return (0);
 438         }
 439 }
 440
 441 int
 442 __bt_fd(dbp)
 443         const DB *dbp;
 444 {
 445         BTREE *t;
 446
 447         t = dbp->internal;
 448
 449         /* Toss any page pinned across calls. */
 450         if (t->bt_pinned != NULL) {
 451                 mpool_put(t->bt_mp, t->bt_pinned, 0);
 452                 t->bt_pinned = NULL;
 453         }
 454
 455         /* In-memory database can't have a file descriptor. */
 456         if (ISSET(t, B_INMEM)) {
 457                 errno = ENOENT;
 458                 return (-1);
 459         }
 460         return (t->bt_fd);
 461 }