lib/libc/db/btree/bt_open.c

   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  * $FreeBSD: src/lib/libc/db/btree/bt_open.c,v 1.7.2.1 2000/11/02 10:30:07 kris Exp $
  33  * $DragonFly: src/lib/libc/db/btree/bt_open.c,v 1.6 2005/11/12 23:01:54 swildner Exp $
  34  *
  35  * @(#)bt_open.c        8.10 (Berkeley) 8/17/94
  36  */
  37
  38 /*
  39  * Implementation of btree access method for 4.4BSD.
  40  *
  41  * The design here was originally based on that of the btree access method
  42  * used in the Postgres database system at UC Berkeley.  This implementation
  43  * is wholly independent of the Postgres code.
  44  */
  45
  46 #include "namespace.h"
  47 #include <sys/param.h>
  48 #include <sys/stat.h>
  49
  50 #include <errno.h>
  51 #include <fcntl.h>
  52 #include <limits.h>
  53 #include <signal.h>
  54 #include <stdio.h>
  55 #include <stdlib.h>
  56 #include <string.h>
  57 #include <unistd.h>
  58 #include "un-namespace.h"
  59
  60 #include <db.h>
  61 #include "btree.h"
  62
  63 #ifdef DEBUG
  64 #undef  MINPSIZE
  65 #define MINPSIZE        128
  66 #endif
  67
  68 static int byteorder (void);
  69 static int nroot (BTREE *);
  70 static int tmp (void);
  71
  72 /*
  73  * __BT_OPEN -- Open a btree.
  74  *
  75  * Creates and fills a DB struct, and calls the routine that actually
  76  * opens the btree.
  77  *
  78  * Parameters:
  79  *      fname:  filename (NULL for in-memory trees)
  80  *      flags:  open flag bits
  81  *      mode:   open permission bits
  82  *      b:      BTREEINFO pointer
  83  *
  84  * Returns:
  85  *      NULL on failure, pointer to DB on success.
  86  *
  87  */
  88 DB *
  89 __bt_open(const char *fname, int flags, int mode, const BTREEINFO *openinfo,
  90           int dflags)
  91 {
  92         struct stat sb;
  93         BTMETA m;
  94         BTREE *t;
  95         BTREEINFO b;
  96         DB *dbp;
  97         pgno_t ncache;
  98         ssize_t nr;
  99         int machine_lorder;
 100
 101         t = NULL;
 102
 103         /*
 104          * Intention is to make sure all of the user's selections are okay
 105          * here and then use them without checking.  Can't be complete, since
 106          * we don't know the right page size, lorder or flags until the backing
 107          * file is opened.  Also, the file's page size can cause the cachesize
 108          * to change.
 109          */
 110         machine_lorder = byteorder();
 111         if (openinfo) {
 112                 b = *openinfo;
 113
 114                 /* Flags: R_DUP. */
 115                 if (b.flags & ~(R_DUP))
 116                         goto einval;
 117
 118                 /*
 119                  * Page size must be indx_t aligned and >= MINPSIZE.  Default
 120                  * page size is set farther on, based on the underlying file
 121                  * transfer size.
 122                  */
 123                 if (b.psize &&
 124                     (b.psize < MINPSIZE || b.psize > MAX_PAGE_OFFSET + 1 ||
 125                     b.psize & (sizeof(indx_t) - 1) ))
 126                         goto einval;
 127
 128                 /* Minimum number of keys per page; absolute minimum is 2. */
 129                 if (b.minkeypage) {
 130                         if (b.minkeypage < 2)
 131                                 goto einval;
 132                 } else
 133                         b.minkeypage = DEFMINKEYPAGE;
 134
 135                 /* If no comparison, use default comparison and prefix. */
 136                 if (b.compare == NULL) {
 137                         b.compare = __bt_defcmp;
 138                         if (b.prefix == NULL)
 139                                 b.prefix = __bt_defpfx;
 140                 }
 141
 142                 if (b.lorder == 0)
 143                         b.lorder = machine_lorder;
 144         } else {
 145                 b.compare = __bt_defcmp;
 146                 b.cachesize = 0;
 147                 b.flags = 0;
 148                 b.lorder = machine_lorder;
 149                 b.minkeypage = DEFMINKEYPAGE;
 150                 b.prefix = __bt_defpfx;
 151                 b.psize = 0;
 152         }
 153
 154         /* Check for the ubiquitous PDP-11. */
 155         if (b.lorder != BIG_ENDIAN && b.lorder != LITTLE_ENDIAN)
 156                 goto einval;
 157
 158         /* Allocate and initialize DB and BTREE structures. */
 159         if ((t = (BTREE *)malloc(sizeof(BTREE))) == NULL)
 160                 goto err;
 161         memset(t, 0, sizeof(BTREE));
 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 *)malloc(sizeof(DB))) == NULL)
 170                 goto err;
 171         memset(t->bt_dbp, 0, sizeof(DB));
 172         if (t->bt_lorder != machine_lorder)
 173                 F_SET(t, B_NEEDSWAP);
 174
 175         dbp->type = DB_BTREE;
 176         dbp->internal = t;
 177         dbp->close = __bt_close;
 178         dbp->del = __bt_delete;
 179         dbp->fd = __bt_fd;
 180         dbp->get = __bt_get;
 181         dbp->put = __bt_put;
 182         dbp->seq = __bt_seq;
 183         dbp->sync = __bt_sync;
 184
 185         /*
 186          * If no file name was supplied, this is an in-memory btree and we
 187          * open a backing temporary file.  Otherwise, it's a disk-based tree.
 188          */
 189         if (fname) {
 190                 switch (flags & O_ACCMODE) {
 191                 case O_RDONLY:
 192                         F_SET(t, B_RDONLY);
 193                         break;
 194                 case O_RDWR:
 195                         break;
 196                 case O_WRONLY:
 197                 default:
 198                         goto einval;
 199                 }
 200
 201                 if ((t->bt_fd = _open(fname, flags, mode)) < 0)
 202                         goto err;
 203
 204         } else {
 205                 if ((flags & O_ACCMODE) != O_RDWR)
 206                         goto einval;
 207                 if ((t->bt_fd = tmp()) == -1)
 208                         goto err;
 209                 F_SET(t, B_INMEM);
 210         }
 211
 212         if (_fcntl(t->bt_fd, F_SETFD, 1) == -1)
 213                 goto err;
 214
 215         if (_fstat(t->bt_fd, &sb))
 216                 goto err;
 217         if (sb.st_size) {
 218                 if ((nr = _read(t->bt_fd, &m, sizeof(BTMETA))) < 0)
 219                         goto err;
 220                 if (nr != sizeof(BTMETA))
 221                         goto eftype;
 222
 223                 /*
 224                  * Read in the meta-data.  This can change the notion of what
 225                  * the lorder, page size and flags are, and, when the page size
 226                  * changes, the cachesize value can change too.  If the user
 227                  * specified the wrong byte order for an existing database, we
 228                  * don't bother to return an error, we just clear the NEEDSWAP
 229                  * bit.
 230                  */
 231                 if (m.magic == BTREEMAGIC)
 232                         F_CLR(t, B_NEEDSWAP);
 233                 else {
 234                         F_SET(t, B_NEEDSWAP);
 235                         M_32_SWAP(m.magic);
 236                         M_32_SWAP(m.version);
 237                         M_32_SWAP(m.psize);
 238                         M_32_SWAP(m.free);
 239                         M_32_SWAP(m.nrecs);
 240                         M_32_SWAP(m.flags);
 241                 }
 242                 if (m.magic != BTREEMAGIC || m.version != BTREEVERSION)
 243                         goto eftype;
 244                 if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 ||
 245                     m.psize & (sizeof(indx_t) - 1) )
 246                         goto eftype;
 247                 if (m.flags & ~SAVEMETA)
 248                         goto eftype;
 249                 b.psize = m.psize;
 250                 F_SET(t, m.flags);
 251                 t->bt_free = m.free;
 252                 t->bt_nrecs = m.nrecs;
 253         } else {
 254                 /*
 255                  * Set the page size to the best value for I/O to this file.
 256                  * Don't overflow the page offset type.
 257                  */
 258                 if (b.psize == 0) {
 259                         b.psize = sb.st_blksize;
 260                         if (b.psize < MINPSIZE)
 261                                 b.psize = MINPSIZE;
 262                         if (b.psize > MAX_PAGE_OFFSET + 1)
 263                                 b.psize = MAX_PAGE_OFFSET + 1;
 264                 }
 265
 266                 /* Set flag if duplicates permitted. */
 267                 if (!(b.flags & R_DUP))
 268                         F_SET(t, B_NODUPS);
 269
 270                 t->bt_free = P_INVALID;
 271                 t->bt_nrecs = 0;
 272                 F_SET(t, B_METADIRTY);
 273         }
 274
 275         t->bt_psize = b.psize;
 276
 277         /* Set the cache size; must be a multiple of the page size. */
 278         if (b.cachesize && b.cachesize & (b.psize - 1) )
 279                 b.cachesize += (~b.cachesize & (b.psize - 1) ) + 1;
 280         if (b.cachesize < b.psize * MINCACHE)
 281                 b.cachesize = b.psize * MINCACHE;
 282
 283         /* Calculate number of pages to cache. */
 284         ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
 285
 286         /*
 287          * The btree data structure requires that at least two keys can fit on
 288          * a page, but other than that there's no fixed requirement.  The user
 289          * specified a minimum number per page, and we translated that into the
 290          * number of bytes a key/data pair can use before being placed on an
 291          * overflow page.  This calculation includes the page header, the size
 292          * of the index referencing the leaf item and the size of the leaf item
 293          * structure.  Also, don't let the user specify a minkeypage such that
 294          * a key/data pair won't fit even if both key and data are on overflow
 295          * pages.
 296          */
 297         t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage -
 298             (sizeof(indx_t) + NBLEAFDBT(0, 0));
 299         if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t))
 300                 t->bt_ovflsize =
 301                     NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
 302
 303         /* Initialize the buffer pool. */
 304         if ((t->bt_mp =
 305             mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
 306                 goto err;
 307         if (!F_ISSET(t, B_INMEM))
 308                 mpool_filter(t->bt_mp, __bt_pgin, __bt_pgout, t);
 309
 310         /* Create a root page if new tree. */
 311         if (nroot(t) == RET_ERROR)
 312                 goto err;
 313
 314         /* Global flags. */
 315         if (dflags & DB_LOCK)
 316                 F_SET(t, B_DB_LOCK);
 317         if (dflags & DB_SHMEM)
 318                 F_SET(t, B_DB_SHMEM);
 319         if (dflags & DB_TXN)
 320                 F_SET(t, B_DB_TXN);
 321
 322         return (dbp);
 323
 324 einval: errno = EINVAL;
 325         goto err;
 326
 327 eftype: errno = EFTYPE;
 328         goto err;
 329
 330 err:    if (t) {
 331                 if (t->bt_dbp)
 332                         free(t->bt_dbp);
 333                 if (t->bt_fd != -1)
 334                         _close(t->bt_fd);
 335                 free(t);
 336         }
 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 ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) {
 356                 mpool_put(t->bt_mp, meta, 0);
 357                 return (RET_SUCCESS);
 358         }
 359         if (errno != EINVAL)            /* It's OK to not exist. */
 360                 return (RET_ERROR);
 361         errno = 0;
 362
 363         if ((meta = mpool_new(t->bt_mp, &npg)) == NULL)
 364                 return (RET_ERROR);
 365
 366         if ((root = mpool_new(t->bt_mp, &npg)) == NULL)
 367                 return (RET_ERROR);
 368
 369         if (npg != P_ROOT)
 370                 return (RET_ERROR);
 371         root->pgno = npg;
 372         root->prevpg = root->nextpg = P_INVALID;
 373         root->lower = BTDATAOFF;
 374         root->upper = t->bt_psize;
 375         root->flags = P_BLEAF;
 376         memset(meta, 0, t->bt_psize);
 377         mpool_put(t->bt_mp, meta, MPOOL_DIRTY);
 378         mpool_put(t->bt_mp, root, MPOOL_DIRTY);
 379         return (RET_SUCCESS);
 380 }
 381
 382 static int
 383 tmp(void)
 384 {
 385         sigset_t set, oset;
 386         int fd;
 387         char *envtmp = NULL;
 388         char path[MAXPATHLEN];
 389
 390         if (issetugid() == 0)
 391                 envtmp = getenv("TMPDIR");
 392         snprintf(path,
 393             sizeof(path), "%s/bt.XXXXXXXXXX", envtmp ? envtmp : "/tmp");
 394
 395         sigfillset(&set);
 396         _sigprocmask(SIG_BLOCK, &set, &oset);
 397         if ((fd = mkstemp(path)) != -1)
 398                 unlink(path);
 399         _sigprocmask(SIG_SETMASK, &oset, NULL);
 400         return(fd);
 401 }
 402
 403 static int
 404 byteorder(void)
 405 {
 406         u_int32_t x;
 407         u_char *p;
 408
 409         x = 0x01020304;
 410         p = (u_char *)&x;
 411         switch (*p) {
 412         case 1:
 413                 return (BIG_ENDIAN);
 414         case 4:
 415                 return (LITTLE_ENDIAN);
 416         default:
 417                 return (0);
 418         }
 419 }
 420
 421 int
 422 __bt_fd(const DB *dbp)
 423 {
 424         BTREE *t;
 425
 426         t = dbp->internal;
 427
 428         /* Toss any page pinned across calls. */
 429         if (t->bt_pinned != NULL) {
 430                 mpool_put(t->bt_mp, t->bt_pinned, 0);
 431                 t->bt_pinned = NULL;
 432         }
 433
 434         /* In-memory database can't have a file descriptor. */
 435         if (F_ISSET(t, B_INMEM)) {
 436                 errno = ENOENT;
 437                 return (-1);
 438         }
 439         return (t->bt_fd);
 440 }