Asterisk - The Open Source Telephony Project  18.5.0
bt_open.c
Go to the documentation of this file.
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. All advertising materials mentioning features or use of this software
17  * must display the following acknowledgement:
18  * This product includes software developed by the University of
19  * California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  * may be used to endorse or promote products derived from this software
22  * without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  */
36 
37 #if defined(LIBC_SCCS) && !defined(lint)
38 static char sccsid[] = "@(#)bt_open.c 8.10 (Berkeley) 8/17/94";
39 #endif /* LIBC_SCCS and not lint */
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 <sys/param.h>
50 #include <sys/stat.h>
51 
52 #include <errno.h>
53 #include <fcntl.h>
54 #include <limits.h>
55 #include <signal.h>
56 #include <stdio.h>
57 #include <stdlib.h>
58 #include <string.h>
59 #include <unistd.h>
60 
61 #include "../include/db.h"
62 #include "btree.h"
63 
64 #ifdef DEBUG
65 #undef MINPSIZE
66 #define MINPSIZE 128
67 #endif
68 
69 static int byteorder __P((void));
70 static int nroot __P((BTREE *));
71 static int tmp __P((void));
72 
73 /*
74  * __BT_OPEN -- Open a btree.
75  *
76  * Creates and fills a DB struct, and calls the routine that actually
77  * opens the btree.
78  *
79  * Parameters:
80  * fname: filename (NULL for in-memory trees)
81  * flags: open flag bits
82  * mode: open permission bits
83  * b: BTREEINFO pointer
84  *
85  * Returns:
86  * NULL on failure, pointer to DB on success.
87  *
88  */
89 DB *
90 __bt_open(fname, flags, mode, openinfo, dflags)
91  const char *fname;
92  int flags, mode, dflags;
93  const BTREEINFO *openinfo;
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;
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
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;
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 *)malloc(sizeof(BTREE))) == NULL)
163  goto err;
164  memset(t, 0, sizeof(BTREE));
165  t->bt_fd = -1; /* Don't close unopened fd on error. */
166  t->bt_lorder = b.lorder;
167  t->bt_order = NOT;
168  t->bt_cmp = b.compare;
169  t->bt_pfx = b.prefix;
170  t->bt_rfd = -1;
171 
172  if ((t->bt_dbp = dbp = (DB *)malloc(sizeof(DB))) == NULL)
173  goto err;
174  memset(t->bt_dbp, 0, sizeof(DB));
175  if (t->bt_lorder != machine_lorder)
176  F_SET(t, B_NEEDSWAP);
177 
178  dbp->type = DB_BTREE;
179  dbp->internal = t;
180  dbp->close = __bt_close;
181  dbp->del = __bt_delete;
182  dbp->fd = __bt_fd;
183  dbp->get = __bt_get;
184  dbp->put = __bt_put;
185  dbp->seq = __bt_seq;
186  dbp->sync = __bt_sync;
187 
188  /*
189  * If no file name was supplied, this is an in-memory btree and we
190  * open a backing temporary file. Otherwise, it's a disk-based tree.
191  */
192  if (fname) {
193  switch (flags & O_ACCMODE) {
194  case O_RDONLY:
195  F_SET(t, B_RDONLY);
196  break;
197  case O_RDWR:
198  break;
199  case O_WRONLY:
200  default:
201  goto einval;
202  }
203 
204  if ((t->bt_fd = open(fname, flags, mode)) < 0)
205  goto err;
206 
207  } else {
208  if ((flags & O_ACCMODE) != O_RDWR)
209  goto einval;
210  if ((t->bt_fd = tmp()) == -1)
211  goto err;
212  F_SET(t, B_INMEM);
213  }
214 
215  if (fcntl(t->bt_fd, F_SETFD, 1) == -1)
216  goto err;
217 
218  if (fstat(t->bt_fd, &sb))
219  goto err;
220  if (sb.st_size) {
221  if ((nr = read(t->bt_fd, &m, sizeof(BTMETA))) < 0)
222  goto err;
223  if (nr != sizeof(BTMETA))
224  goto eftype;
225 
226  /*
227  * Read in the meta-data. This can change the notion of what
228  * the lorder, page size and flags are, and, when the page size
229  * changes, the cachesize value can change too. If the user
230  * specified the wrong byte order for an existing database, we
231  * don't bother to return an error, we just clear the NEEDSWAP
232  * bit.
233  */
234  if (m.magic == BTREEMAGIC)
235  F_CLR(t, B_NEEDSWAP);
236  else {
237  F_SET(t, B_NEEDSWAP);
238  M_32_SWAP(m.magic);
239  M_32_SWAP(m.version);
240  M_32_SWAP(m.psize);
241  M_32_SWAP(m.free);
242  M_32_SWAP(m.nrecs);
243  M_32_SWAP(m.flags);
244  }
245  if (m.magic != BTREEMAGIC || m.version != BTREEVERSION)
246  goto eftype;
247  if (m.psize < MINPSIZE || m.psize > MAX_PAGE_OFFSET + 1 ||
248  m.psize & (sizeof(indx_t) - 1))
249  goto eftype;
250  if (m.flags & ~SAVEMETA)
251  goto eftype;
252  b.psize = m.psize;
253  F_SET(t, m.flags);
254  t->bt_free = m.free;
255  t->bt_nrecs = m.nrecs;
256  } else {
257  /*
258  * Set the page size to the best value for I/O to this file.
259  * Don't overflow the page offset type.
260  */
261  if (b.psize == 0) {
262 #ifdef _STATBUF_ST_BLKSIZE
263  b.psize = sb.st_blksize;
264 #endif
265  if (b.psize < MINPSIZE)
266  b.psize = MINPSIZE;
267  if (b.psize > MAX_PAGE_OFFSET + 1)
268  b.psize = MAX_PAGE_OFFSET + 1;
269  }
270 
271  /* Set flag if duplicates permitted. */
272  if (!(b.flags & R_DUP))
273  F_SET(t, B_NODUPS);
274 
275  t->bt_free = P_INVALID;
276  t->bt_nrecs = 0;
277  F_SET(t, B_METADIRTY);
278  }
279 
280  t->bt_psize = b.psize;
281 
282  /* Set the cache size; must be a multiple of the page size. */
283  if (b.cachesize && b.cachesize & (b.psize - 1))
284  b.cachesize += (~b.cachesize & (b.psize - 1)) + 1;
285  if (b.cachesize < b.psize * MINCACHE)
286  b.cachesize = b.psize * MINCACHE;
287 
288  /* Calculate number of pages to cache. */
289  ncache = (b.cachesize + t->bt_psize - 1) / t->bt_psize;
290 
291  /*
292  * The btree data structure requires that at least two keys can fit on
293  * a page, but other than that there's no fixed requirement. The user
294  * specified a minimum number per page, and we translated that into the
295  * number of bytes a key/data pair can use before being placed on an
296  * overflow page. This calculation includes the page header, the size
297  * of the index referencing the leaf item and the size of the leaf item
298  * structure. Also, don't let the user specify a minkeypage such that
299  * a key/data pair won't fit even if both key and data are on overflow
300  * pages.
301  */
302  t->bt_ovflsize = (t->bt_psize - BTDATAOFF) / b.minkeypage -
303  (sizeof(indx_t) + NBLEAFDBT(0, 0));
304  if (t->bt_ovflsize < NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t))
305  t->bt_ovflsize =
306  NBLEAFDBT(NOVFLSIZE, NOVFLSIZE) + sizeof(indx_t);
307 
308  /* Initialize the buffer pool. */
309  if ((t->bt_mp =
310  mpool_open(NULL, t->bt_fd, t->bt_psize, ncache)) == NULL)
311  goto err;
312  if (!F_ISSET(t, B_INMEM))
314 
315  /* Create a root page if new tree. */
316  if (nroot(t) == RET_ERROR)
317  goto err;
318 
319  /* Global flags. */
320  if (dflags & DB_LOCK)
321  F_SET(t, B_DB_LOCK);
322  if (dflags & DB_SHMEM)
323  F_SET(t, B_DB_SHMEM);
324  if (dflags & DB_TXN)
325  F_SET(t, B_DB_TXN);
326 
327  return (dbp);
328 
329 einval: errno = EINVAL;
330  goto err;
331 
332 eftype: errno = EFTYPE;
333  goto err;
334 
335 err: if (t) {
336  if (t->bt_dbp)
337  free(t->bt_dbp);
338  if (t->bt_fd != -1)
339  (void)close(t->bt_fd);
340  free(t);
341  }
342  return (NULL);
343 }
344 
345 /*
346  * NROOT -- Create the root of a new tree.
347  *
348  * Parameters:
349  * t: tree
350  *
351  * Returns:
352  * RET_ERROR, RET_SUCCESS
353  */
354 static int
356  BTREE *t;
357 {
358  PAGE *meta, *root;
359  pgno_t npg;
360 
361  if ((meta = mpool_get(t->bt_mp, 0, 0)) != NULL) {
362  mpool_put(t->bt_mp, meta, 0);
363  return (RET_SUCCESS);
364  }
365  if (errno != EINVAL) /* It's OK to not exist. */
366  return (RET_ERROR);
367  errno = 0;
368 
369  if ((meta = mpool_new(t->bt_mp, &npg)) == NULL)
370  return (RET_ERROR);
371 
372  if ((root = mpool_new(t->bt_mp, &npg)) == NULL)
373  return (RET_ERROR);
374 
375  if (npg != P_ROOT)
376  return (RET_ERROR);
377  root->pgno = npg;
378  root->prevpg = root->nextpg = P_INVALID;
379  root->lower = BTDATAOFF;
380  root->upper = t->bt_psize;
381  root->flags = P_BLEAF;
382  memset(meta, 0, t->bt_psize);
383  mpool_put(t->bt_mp, meta, MPOOL_DIRTY);
384  mpool_put(t->bt_mp, root, MPOOL_DIRTY);
385  return (RET_SUCCESS);
386 }
387 
388 static int
390 {
391  sigset_t set, oset;
392  int fd;
393  const char *envtmp;
394  char *path;
395  static const char fmt[] = "%s/bt.XXXXXX";
396  size_t n;
397 
398  envtmp = getenv("TMPDIR");
399  if (!envtmp)
400  envtmp = "/tmp";
401  n = strlen (envtmp) + sizeof fmt;
402 #ifdef __GNUC__
403  path = __builtin_alloca(n);
404 #else
405  path = malloc(n);
406 #endif
407  (void)snprintf(path, n, fmt, envtmp ? envtmp : "/tmp");
408 
409  (void)sigfillset(&set);
410  (void)sigprocmask(SIG_BLOCK, &set, &oset);
411  if ((fd = mkstemp(path)) != -1)
412  (void)unlink(path);
413  (void)sigprocmask(SIG_SETMASK, &oset, NULL);
414 #ifndef __GNUC__
415  free(path);
416 #endif
417  return(fd);
418 }
419 
420 static int
422 {
423  u_int32_t x;
424  u_char *p;
425 
426  x = 0x01020304;
427  p = (u_char *)&x;
428  switch (*p) {
429  case 1:
430  return (BIG_ENDIAN);
431  case 4:
432  return (LITTLE_ENDIAN);
433  default:
434  return (0);
435  }
436 }
437 
438 int
440  const DB *dbp;
441 {
442  BTREE *t;
443 
444  t = dbp->internal;
445 
446  /* Toss any page pinned across calls. */
447  if (t->bt_pinned != NULL) {
448  mpool_put(t->bt_mp, t->bt_pinned, 0);
449  t->bt_pinned = NULL;
450  }
451 
452  /* In-memory database can't have a file descriptor. */
453  if (F_ISSET(t, B_INMEM)) {
454  errno = ENOENT;
455  return (-1);
456  }
457  return (t->bt_fd);
458 }
int __bt_close(DB *dbp)
Definition: bt_close.c:64
int bt_rfd
Definition: btree.h:353
Definition: btree.h:75
#define MINCACHE
Definition: btree.h:55
#define B_DB_TXN
Definition: btree.h:387
#define RET_ERROR
Definition: db.h:51
#define B_DB_SHMEM
Definition: btree.h:386
pgno_t pgno
Definition: btree.h:76
#define F_SET(p, f)
Definition: btree.h:40
#define F_ISSET(p, f)
Definition: btree.h:42
static int nroot(BTREE *t)
Definition: bt_open.c:355
int mpool_put(MPOOL *mp, void *page, u_int flags)
Definition: mpool.c:251
#define DB_TXN
Definition: db.h:125
#define BTREEVERSION
Definition: db.h:142
static int tmp()
Definition: bt_open.c:389
dflags
Definition: app_dictate.c:63
#define B_RDONLY
Definition: btree.h:372
int __bt_put(DB *dbp, DBT *key, const DBT *data, u_int flags) const
Definition: bt_put.c:67
u_int32_t flags
Definition: btree.h:388
void * internal
Definition: db.h:137
#define F_CLR(p, f)
Definition: btree.h:41
indx_t lower
Definition: btree.h:89
#define NULL
Definition: resample.c:96
#define MINPSIZE
Definition: btree.h:56
#define B_NEEDSWAP
Definition: btree.h:371
u_int16_t indx_t
Definition: db.h:80
indx_t upper
Definition: btree.h:90
recno_t bt_nrecs
Definition: btree.h:360
Definition: btree.h:312
#define NOVFLSIZE
Definition: btree.h:117
Definition: db.h:145
#define MPOOL_DIRTY
Definition: mpool.h:61
meta
below block is needed for mssql
Definition: env.py:22
u_long flags
Definition: db.h:147
char * malloc()
void * mpool_get(MPOOL *mp, pgno_t pgno, u_int flags)
Definition: mpool.c:165
void free()
void __bt_pgout(void *t, pgno_t pg, void *pp)
Definition: bt_conv.c:129
u_int32_t flags
Definition: btree.h:87
#define SAVEMETA
Definition: btree.h:307
u_int cachesize
Definition: db.h:148
Definition: db.h:103
#define RET_SUCCESS
Definition: db.h:52
PAGE * bt_pinned
Definition: btree.h:318
#define DB_LOCK
Definition: db.h:123
#define BIG_ENDIAN
void mpool_filter(MPOOL *mp, void *pgin, void *pgout, void *pgcookie)
Definition: mpool.c:114
void __bt_pgin(void *t, pgno_t pg, void *pp)
Definition: bt_conv.c:61
int bt_lorder
Definition: btree.h:340
#define MAX_PAGE_OFFSET
Definition: db.h:79
DB * __bt_open(char *fname, int flags, int mode, const BTREEINFO *openinfo, int dflags) const
Definition: bt_open.c:90
MPOOL * bt_mp
Definition: btree.h:313
DBTYPE type
Definition: db.h:130
int __bt_defcmp(DBT *a, DBT *b) const
Definition: bt_utils.c:216
static DB * dbp
Definition: hsearch.c:49
#define B_NODUPS
Definition: btree.h:374
u_int32_t pgno_t
Definition: db.h:78
pgno_t nextpg
Definition: btree.h:78
int errno
pgno_t prevpg
Definition: btree.h:77
#define R_DUP
Definition: db.h:146
#define LITTLE_ENDIAN
#define NBLEAFDBT(ksize, dsize)
Definition: btree.h:195
#define DB_SHMEM
Definition: db.h:124
u_int32_t bt_psize
Definition: btree.h:338
#define EFTYPE
#define P_ROOT
Definition: btree.h:65
#define DEFMINKEYPAGE
Definition: btree.h:54
unsigned int u_int32_t
int lorder
Definition: db.h:156
enum _btree::@518 bt_order
static int byteorder __P((void))
#define B_METADIRTY
Definition: btree.h:369
int minkeypage
Definition: db.h:150
static int byteorder()
Definition: bt_open.c:421
int __bt_delete(DB *dbp, const DBT *key, u_int flags) const
Definition: bt_delete.c:63
int __bt_sync(DB *dbp, u_int flags) const
Definition: bt_close.c:119
#define B_INMEM
Definition: btree.h:368
#define P_INVALID
Definition: btree.h:63
#define BTREEMAGIC
Definition: db.h:141
#define BTDATAOFF
Definition: btree.h:95
int __bt_get(DB *dbp, const DBT *key, DBT *data, u_int flags) const
Definition: bt_get.c:63
int bt_fd
Definition: btree.h:335
void * mpool_new(MPOOL *mp, pgno_t *pgnoaddr)
Definition: mpool.c:130
int __bt_seq(DB *dbp, DBT *key, DBT *data, u_int flags) const
Definition: bt_seq.c:77
indx_t bt_ovflsize
Definition: btree.h:339
DB * bt_dbp
Definition: btree.h:315
Definition: db.h:129
static struct test_val b
u_int psize
Definition: db.h:151
MPOOL * mpool_open(void *key, int fd, pgno_t pagesize, pgno_t maxcache)
Definition: mpool.c:74
#define P_BLEAF
Definition: btree.h:81
pgno_t bt_free
Definition: btree.h:337
#define B_DB_LOCK
Definition: btree.h:385
size_t __bt_defpfx(DBT *a, DBT *b) const
Definition: bt_utils.c:246
Definition: btree.h:300
int __bt_fd(DB *dbp) const
Definition: bt_open.c:439