xref: /titanic_44/usr/src/cmd/sendmail/db/db/db.c (revision 7c478bd95313f5f23a4c958a745db2134aa03244)
1 /*-
2  * See the file LICENSE for redistribution information.
3  *
4  * Copyright (c) 1996, 1997, 1998
5  *	Sleepycat Software.  All rights reserved.
6  */
7 /*
8  * Copyright (c) 1990, 1993, 1994, 1995, 1996
9  *	Keith Bostic.  All rights reserved.
10  */
11 /*
12  * Copyright (c) 1990, 1993, 1994, 1995
13  *	The Regents of the University of California.  All rights reserved.
14  *
15  * Redistribution and use in source and binary forms, with or without
16  * modification, are permitted provided that the following conditions
17  * are met:
18  * 1. Redistributions of source code must retain the above copyright
19  *    notice, this list of conditions and the following disclaimer.
20  * 2. Redistributions in binary form must reproduce the above copyright
21  *    notice, this list of conditions and the following disclaimer in the
22  *    documentation and/or other materials provided with the distribution.
23  * 3. All advertising materials mentioning features or use of this software
24  *    must display the following acknowledgement:
25  *	This product includes software developed by the University of
26  *	California, Berkeley and its contributors.
27  * 4. Neither the name of the University nor the names of its contributors
28  *    may be used to endorse or promote products derived from this software
29  *    without specific prior written permission.
30  *
31  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
32  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
33  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
34  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
35  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
36  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
37  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
38  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
39  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
40  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41  * SUCH DAMAGE.
42  */
43 
44 #include "config.h"
45 
46 #ifndef lint
47 static const char sccsid[] = "@(#)db.c	10.75 (Sleepycat) 12/3/98";
48 #endif /* not lint */
49 
50 #ifndef NO_SYSTEM_INCLUDES
51 #include <sys/types.h>
52 
53 #include <errno.h>
54 #include <stddef.h>
55 #include <stdlib.h>
56 #include <string.h>
57 #endif
58 
59 #include "db_int.h"
60 #include "shqueue.h"
61 #include "db_page.h"
62 #include "db_shash.h"
63 #include "db_swap.h"
64 #include "btree.h"
65 #include "hash.h"
66 #include "mp.h"
67 #include "db_am.h"
68 #include "common_ext.h"
69 
70 /*
71  * If the metadata page has the flag set, set the local flag.  If the page
72  * does NOT have the flag set, return EINVAL if the user's dbinfo argument
73  * caused us to already set the local flag.
74  */
75 #define	DBINFO_FCHK(dbp, fn, meta_flags, m_name, dbp_name) {		\
76 	if ((meta_flags) & (m_name))					\
77 		F_SET(dbp, dbp_name);					\
78 	else								\
79 		if (F_ISSET(dbp, dbp_name)) {				\
80 			__db_err(dbenv,					\
81 	    "%s: %s specified in dbinfo argument but not set in file",	\
82 			    fname, fn);					\
83 			goto einval;					\
84 		}							\
85 }
86 
87 /*
88  * db_open --
89  *	Main library interface to the DB access methods.
90  */
91 int
db_open(fname,type,flags,mode,dbenv,dbinfo,dbpp)92 db_open(fname, type, flags, mode, dbenv, dbinfo, dbpp)
93 	const char *fname;
94 	DBTYPE type;
95 	u_int32_t flags;
96 	int mode;
97 	DB_ENV *dbenv;
98 	DB_INFO *dbinfo;
99 	DB **dbpp;
100 {
101 	BTMETA *btm;
102 	DB *dbp;
103 	DBT pgcookie;
104 	DB_ENV *envp, t_dbenv;
105 	DB_MPOOL_FINFO finfo;
106 	DB_PGINFO pginfo;
107 	HASHHDR *hashm;
108 	size_t cachesize;
109 	ssize_t nr;
110 	u_int32_t iopsize;
111 	int fd, ftype, need_fileid, restore, ret, retry_cnt, swapped;
112 	char *real_name, mbuf[512];
113 
114 	/* Validate arguments. */
115 #ifdef HAVE_SPINLOCKS
116 #define	OKFLAGS	(DB_CREATE | DB_FCNTL_LOCKING | DB_NOMMAP | DB_RDONLY | DB_THREAD | DB_TRUNCATE)
117 #else
118 #define	OKFLAGS	(DB_CREATE | DB_FCNTL_LOCKING | DB_NOMMAP | DB_RDONLY | DB_TRUNCATE)
119 #endif
120 	if ((ret = __db_fchk(dbenv, "db_open", flags, OKFLAGS)) != 0)
121 		return (ret);
122 
123 	if (dbenv != NULL) {
124 		/*
125 		 * You can't specify threads during the db_open() if the
126 		 * environment wasn't configured with them.
127 		 */
128 		if (LF_ISSET(DB_THREAD) && !F_ISSET(dbenv, DB_ENV_THREAD)) {
129 			__db_err(dbenv,
130 			    "environment not created using DB_THREAD");
131 			return (EINVAL);
132 		}
133 
134 		/*
135 		 * Specifying a cachesize to db_open(3), after creating an
136 		 * environment with DB_INIT_MPOOL, is a common mistake.
137 		 */
138 		if (dbenv->mp_info != NULL &&
139 		    dbinfo != NULL && dbinfo->db_cachesize != 0) {
140 			__db_err(dbenv,
141 			    "cachesize will be ignored if environment exists");
142 			return (EINVAL);
143 		}
144 	}
145 
146 	/* Allocate the DB structure, reference the DB_ENV structure. */
147 	if ((ret = __os_calloc(1, sizeof(DB), &dbp)) != 0)
148 		return (ret);
149 	dbp->dbenv = dbenv;
150 
151 	/* Initialize for error return. */
152 	dbp->saved_open_fd = fd = -1;
153 	need_fileid = 1;
154 	real_name = NULL;
155 
156 	/* Random initialization. */
157 	TAILQ_INIT(&dbp->free_queue);
158 	TAILQ_INIT(&dbp->active_queue);
159 	if ((ret = __db_init_wrapper(dbp)) != 0)
160 		goto err;
161 
162 	/* Convert the db_open(3) flags. */
163 	if (LF_ISSET(DB_RDONLY))
164 		F_SET(dbp, DB_AM_RDONLY);
165 	if (LF_ISSET(DB_THREAD))
166 		F_SET(dbp, DB_AM_THREAD);
167 
168 	/* Convert the dbinfo structure flags. */
169 	if (dbinfo != NULL) {
170 		/*
171 		 * !!!
172 		 * We can't check for illegal flags until we know what type
173 		 * of open we're doing.
174 		 */
175 		if (F_ISSET(dbinfo, DB_DELIMITER))
176 			F_SET(dbp, DB_RE_DELIMITER);
177 		if (F_ISSET(dbinfo, DB_DUP))
178 			F_SET(dbp, DB_AM_DUP);
179 		if (F_ISSET(dbinfo, DB_FIXEDLEN))
180 			F_SET(dbp, DB_RE_FIXEDLEN);
181 		if (F_ISSET(dbinfo, DB_PAD))
182 			F_SET(dbp, DB_RE_PAD);
183 		if (F_ISSET(dbinfo, DB_RECNUM))
184 			F_SET(dbp, DB_BT_RECNUM);
185 		if (F_ISSET(dbinfo, DB_RENUMBER))
186 			F_SET(dbp, DB_RE_RENUMBER);
187 		if (F_ISSET(dbinfo, DB_SNAPSHOT))
188 			F_SET(dbp, DB_RE_SNAPSHOT);
189 	}
190 
191 	/*
192 	 * Set based on the dbenv fields, although no logging or transactions
193 	 * are possible for temporary files.
194 	 */
195 	if (dbenv != NULL) {
196 		if (dbenv->lk_info != NULL)
197 			if (F_ISSET(dbenv, DB_ENV_CDB))
198 				F_SET(dbp, DB_AM_CDB);
199 			else
200 				F_SET(dbp, DB_AM_LOCKING);
201 		if (fname != NULL && dbenv->lg_info != NULL)
202 			F_SET(dbp, DB_AM_LOGGING);
203 	}
204 
205 	/* Set the common fields. */
206 	if (dbinfo == NULL) {
207 		dbp->pgsize = 0;
208 		dbp->db_malloc = NULL;
209 		dbp->dup_compare = NULL;
210 	} else {
211 		/*
212 		 * We don't want anything that's not a power-of-2, as we rely
213 		 * on that for alignment of various types on the pages.
214 		 */
215 		if ((dbp->pgsize = dbinfo->db_pagesize) != 0 &&
216 		    (u_int32_t)1 << __db_log2(dbp->pgsize) != dbp->pgsize) {
217 			__db_err(dbenv, "page sizes must be a power-of-2");
218 			goto einval;
219 		}
220 		dbp->pgsize = dbinfo->db_pagesize;
221 		dbp->db_malloc = dbinfo->db_malloc;
222 		if (F_ISSET(dbinfo, DB_DUPSORT)) {
223 			if (F_ISSET(dbinfo, DB_DUP))
224 				dbp->dup_compare = dbinfo->dup_compare == NULL ?
225 				    __bam_defcmp : dbinfo->dup_compare;
226 			else {
227 				__db_err(dbenv, "DB_DUPSORT requires DB_DUP");
228 				goto einval;
229 			}
230 			F_CLR(dbinfo, DB_DUPSORT);
231 		}
232 	}
233 
234 	/* Fill in the default file mode. */
235 	if (mode == 0)
236 		mode = __db_omode("rwrw--");
237 
238 	/* Check if the user wants us to swap byte order. */
239 	if (dbinfo != NULL)
240 		switch (ret = __db_byteorder(dbenv, dbinfo->db_lorder)) {
241 		case 0:
242 			break;
243 		case DB_SWAPBYTES:
244 			F_SET(dbp, DB_AM_SWAP);
245 			break;
246 		default:
247 			goto err;
248 		}
249 	dbp->byteswapped = F_ISSET(dbp, DB_AM_SWAP) ? 1 : 0;
250 
251 	/*
252 	 * If we have a file name, try and read the first page, figure out
253 	 * what type of file it is, and initialize everything we can based
254 	 * on that file's meta-data page.
255 	 *
256 	 * XXX
257 	 * We don't actually expect zero-length strings as arguments.  We
258 	 * do the check, permitting them, because scripting languages, e.g.,
259 	 * the Tcl test suite, doesn't know anything about passing NULL's.
260 	 */
261 	if (fname != NULL && fname[0] != '\0') {
262 		/* Get the real file name. */
263 		if ((ret = __db_appname(dbenv,
264 		     DB_APP_DATA, NULL, fname, 0, NULL, &real_name)) != 0)
265 			goto err;
266 
267 		/*
268 		 * Open the backing file.  We need to make sure that multiple
269 		 * processes attempting to create the file at the same time
270 		 * are properly ordered so that only one of them creates the
271 		 * "unique" file id, so we open it O_EXCL and O_CREAT so two
272 		 * simultaneous attempts to create the region will return
273 		 * failure in one of the attempts.  If we're one of the ones
274 		 * that fail, we simply retry without the O_CREAT flag, which
275 		 * will require that the meta-data page exist.
276 		 */
277 		retry_cnt = 0;
278 open_retry:	if (LF_ISSET(DB_CREATE)) {
279 			if ((ret = __db_open(real_name, flags | DB_EXCL,
280 			    OKFLAGS | DB_EXCL, mode, &fd)) != 0)
281 				if (ret == EEXIST) {
282 					LF_CLR(DB_CREATE);
283 					goto open_retry;
284 				} else {
285 					__db_err(dbenv,
286 					    "%s: %s", fname, strerror(ret));
287 					goto err;
288 				}
289 		} else
290 			if ((ret = __db_open(real_name,
291 			    flags, OKFLAGS, mode, &fd)) != 0) {
292 				__db_err(dbenv, "%s: %s", fname, strerror(ret));
293 				goto err;
294 			}
295 
296 		/*
297 		 * Use the optimum I/O size as the pagesize if a pagesize not
298 		 * specified.  Some filesystems have 64K as their optimum I/O
299 		 * size, but as that results in impossibly large default cache
300 		 * sizes, we limit the default pagesize to 16K.
301 		 */
302 		if (dbp->pgsize == 0) {
303 			if ((ret = __os_ioinfo(real_name,
304 			    fd, NULL, NULL, &iopsize)) != 0) {
305 				__db_err(dbenv,
306 				    "%s: %s", real_name, strerror(ret));
307 				goto err;
308 			}
309 			if (iopsize < 512)
310 				iopsize = 512;
311 			if (iopsize > 16 * 1024)
312 				iopsize = 16 * 1024;
313 
314 			/*
315 			 * Sheer paranoia, but we don't want anything that's
316 			 * not a power-of-2, as we rely on that for alignment
317 			 * of various types on the pages.
318 			 */
319 			DB_ROUNDOFF(iopsize, 512);
320 
321 			dbp->pgsize = iopsize;
322 			F_SET(dbp, DB_AM_PGDEF);
323 		}
324 
325 		/*
326 		 * Try and read the first disk sector -- this code assumes
327 		 * that the meta-data for all access methods fits in 512
328 		 * bytes, and that no database will be smaller than that.
329 		 */
330 		if ((ret = __os_read(fd, mbuf, sizeof(mbuf), &nr)) != 0)
331 			goto err;
332 
333 		if (LF_ISSET(DB_FCNTL_LOCKING))
334 			dbp->saved_open_fd = fd;
335 		else
336 			(void)__os_close(fd);
337 		fd = -1;
338 
339 		if (nr != sizeof(mbuf)) {
340 			if (nr != 0) {
341 				__db_err(dbenv,
342 				    "%s: unexpected file format", fname);
343 				goto einval;
344 			}
345 			/*
346 			 * The only way we can reach here with the DB_CREATE
347 			 * flag set is if we created the file.  If that's not
348 			 * the case, then a) someone else created the file
349 			 * but has not yet written out the meta-data page, or
350 			 * b) we truncated the file (DB_TRUNCATE) leaving it
351 			 * zero-length.  In the case of a), we want to sleep
352 			 * and give the file creator some time to write the
353 			 * metadata page.  In the case of b), charge forward.
354 			 * Note, there is a race in the case of two processes
355 			 * opening the file with the DB_TRUNCATE flag set at
356 			 * roughly the same time, and they could theoretically
357 			 * hurt each other, although it's pretty unlikely.
358 			 */
359 			if (retry_cnt++ < 3 &&
360 			    !LF_ISSET(DB_CREATE | DB_TRUNCATE)) {
361 				__os_sleep(1, 0);
362 				goto open_retry;
363 			}
364 			if (type == DB_UNKNOWN) {
365 				__db_err(dbenv,
366 				    "%s: DBTYPE of unknown with empty file",
367 				    fname);
368 				goto einval;
369 			}
370 			goto empty;
371 		}
372 
373 		/*
374 		 * A found file overrides some user information.  We'll check
375 		 * for possible error conditions based on conflicts between
376 		 * the file and the user's arguments below.
377 		 */
378 		swapped = 0;
379 		F_CLR(dbp, DB_AM_SWAP);
380 
381 retry:		switch (((BTMETA *)mbuf)->magic) {
382 		case DB_BTREEMAGIC:
383 			if (type != DB_BTREE &&
384 			    type != DB_RECNO && type != DB_UNKNOWN)
385 				goto einval;
386 
387 			btm = (BTMETA *)mbuf;
388 			if (swapped && (ret = __bam_mswap((PAGE *)btm)) != 0)
389 				goto err;
390 
391 			if (btm->version < DB_BTREEOLDVER ||
392 			    btm->version > DB_BTREEVERSION) {
393 				__db_err(dbenv,
394 				    "%s: unsupported btree version number %lu",
395 				    fname, (u_long)btm->version);
396 				goto einval;
397 			}
398 			dbp->pgsize = btm->pagesize;
399 			F_CLR(dbp, DB_AM_PGDEF);
400 
401 			if ((ret = __db_fchk(dbenv,
402 			    "db_open", btm->flags, BTM_MASK)) != 0)
403 				goto err;
404 			DBINFO_FCHK(dbp, "DB_DUP",
405 			    btm->flags, BTM_DUP, DB_AM_DUP);
406 			if (F_ISSET(btm, BTM_RECNO)) {
407 				DBINFO_FCHK(dbp, "DB_FIXEDLEN",
408 				    btm->flags, BTM_FIXEDLEN, DB_RE_FIXEDLEN);
409 				DBINFO_FCHK(dbp, "DB_RENUMBER",
410 				    btm->flags, BTM_RENUMBER, DB_RE_RENUMBER);
411 				type = DB_RECNO;
412 			} else {
413 				DBINFO_FCHK(dbp, "DB_RECNUM",
414 				    btm->flags, BTM_RECNUM, DB_BT_RECNUM);
415 				type = DB_BTREE;
416 			}
417 
418 			/* Copy the file's unique id. */
419 			need_fileid = 0;
420 			memcpy(dbp->fileid, btm->uid, DB_FILE_ID_LEN);
421 			break;
422 		case DB_HASHMAGIC:
423 			if (type != DB_HASH && type != DB_UNKNOWN)
424 				goto einval;
425 
426 			hashm = (HASHHDR *)mbuf;
427 			if (swapped && (ret = __ham_mswap((PAGE *)hashm)) != 0)
428 				goto err;
429 
430 			if (hashm->version < DB_HASHOLDVER ||
431 			    hashm->version > DB_HASHVERSION) {
432 				__db_err(dbenv,
433 				    "%s: unsupported hash version number %lu",
434 				    fname, hashm->version);
435 				goto einval;
436 			}
437 			dbp->pgsize = hashm->pagesize;
438 			F_CLR(dbp, DB_AM_PGDEF);
439 
440 			if ((ret = __db_fchk(dbenv,
441 			    "db_open", hashm->flags, DB_HASH_DUP)) != 0)
442 				goto err;
443 			DBINFO_FCHK(dbp, "DB_DUP",
444 			    hashm->flags, DB_HASH_DUP, DB_AM_DUP);
445 			type = DB_HASH;
446 
447 			/* Copy the file's unique id. */
448 			need_fileid = 0;
449 			memcpy(dbp->fileid, hashm->uid, DB_FILE_ID_LEN);
450 			break;
451 		default:
452 			if (swapped) {
453 				__db_err(dbenv, "unrecognized file type");
454 				goto einval;
455 			}
456 			M_32_SWAP(((BTMETA *)mbuf)->magic);
457 			F_SET(dbp, DB_AM_SWAP);
458 
459 			swapped = 1;
460 			goto retry;
461 		}
462 	} else {
463 		fname = real_name = NULL;
464 
465 		if (type == DB_UNKNOWN) {
466 			__db_err(dbenv,
467 			    "DBTYPE of unknown without existing file");
468 			goto einval;
469 		}
470 		F_SET(dbp, DB_AM_INMEM);
471 	}
472 
473 empty:	/*
474 	 * By the time we get here we've either set the type or we're taking
475 	 * it from the user.
476 	 */
477 	dbp->type = type;
478 
479 	/*
480 	 * Set the page size to the best value for I/O to this file.  Don't
481 	 * overflow the page offset type.  The page size must be db_indx_t
482 	 * aligned and >= MIN_PAGE_SIZE.
483 	 *
484 	 * XXX
485 	 * Should we be checking for a page size that's not a multiple of 512?
486 	 */
487 	if (dbp->pgsize == 0) {
488 		F_SET(dbp, DB_AM_PGDEF);
489 		dbp->pgsize = 8 * 1024;
490 	}
491 	if (dbp->pgsize < DB_MIN_PGSIZE ||
492 	    dbp->pgsize > DB_MAX_PGSIZE ||
493 	    dbp->pgsize & (sizeof(db_indx_t) - 1)) {
494 		__db_err(dbenv, "illegal page size");
495 		goto einval;
496 	}
497 
498 	/*
499 	 * If no mpool supplied by the application, attach to a local,
500 	 * created buffer pool.
501 	 *
502 	 * XXX
503 	 * If the user has a DB_ENV structure, we have to use a temporary
504 	 * one so that we don't step on their values.  If the user doesn't,
505 	 * we have to create one, and keep it around until the call to the
506 	 * memp_close() function.  This is all so the mpool functions get
507 	 * the error stuff right.
508 	 */
509 	if (dbenv == NULL || dbenv->mp_info == NULL) {
510 		F_SET(dbp, DB_AM_MLOCAL);
511 
512 		if (dbenv == NULL) {
513 			if ((ret = __os_calloc(1,
514 			    sizeof(DB_ENV), &dbp->mp_dbenv)) != 0)
515 				goto err;
516 
517 			envp = dbp->mp_dbenv;
518 			restore = 0;
519 		} else {
520 			t_dbenv = *dbenv;
521 
522 			envp = dbenv;
523 			restore = 1;
524 		}
525 
526 		/*
527 		 * Set and/or correct the cache size; must be a multiple of
528 		 * the page size.
529 		 */
530 		if (dbinfo == NULL || dbinfo->db_cachesize == 0)
531 			cachesize = dbp->pgsize * DB_MINCACHE;
532 		else {
533 			cachesize = dbinfo->db_cachesize;
534 			if (cachesize & (dbp->pgsize - 1))
535 				cachesize +=
536 				    (~cachesize & (dbp->pgsize - 1)) + 1;
537 			if (cachesize < dbp->pgsize * DB_MINCACHE)
538 				cachesize = dbp->pgsize * DB_MINCACHE;
539 			if (cachesize < 20 * 1024)
540 				cachesize = 20 * 1024;
541 		}
542 		envp->mp_size = cachesize;
543 
544 		if ((ret = memp_open(NULL, DB_CREATE | DB_MPOOL_PRIVATE |
545 		    (F_ISSET(dbp, DB_AM_THREAD) ? DB_THREAD : 0),
546 		    __db_omode("rw----"), envp, &dbp->mp)) != 0)
547 			goto err;
548 		if (restore)
549 			*dbenv = t_dbenv;
550 	} else
551 		dbp->mp = dbenv->mp_info;
552 
553 	/* Register DB's pgin/pgout functions. */
554 	if ((ret = memp_register(dbp->mp,
555 	    DB_FTYPE_BTREE, __bam_pgin, __bam_pgout)) != 0)
556 		goto err;
557 	if ((ret = memp_register(dbp->mp,
558 	    DB_FTYPE_HASH, __ham_pgin, __ham_pgout)) != 0)
559 		goto err;
560 
561 	/*
562 	 * If we don't already have one, get a unique file ID.  If the file
563 	 * is a temporary file, then we have to create a unique file ID --
564 	 * no backing file will be created until the mpool cache is filled
565 	 * forcing it to go to disk.  The created ID must never match any
566 	 * potential real file ID -- we know it won't because real file IDs
567 	 * contain a time stamp after the dev/ino pair, and we're simply
568 	 * storing a 4-byte locker ID.
569 	 *
570 	 * XXX
571 	 * Store the file id in the locker structure -- we can get it from
572 	 * there as necessary, and it saves having two copies.
573 	 */
574 	if (need_fileid)
575 		if (fname == NULL) {
576 			memset(dbp->fileid, 0, DB_FILE_ID_LEN);
577 			if (F_ISSET(dbp, DB_AM_LOCKING) &&
578 			    (ret = lock_id(dbenv->lk_info,
579 			    (u_int32_t *)dbp->fileid)) != 0)
580 				goto err;
581 		} else
582 			if ((ret = __os_fileid(dbenv,
583 			    real_name, 1, dbp->fileid)) != 0)
584 				goto err;
585 
586 	/* No further use for the real name. */
587 	if (real_name != NULL)
588 		__os_freestr(real_name);
589 	real_name = NULL;
590 
591 	/*
592 	 * Open a backing file in the memory pool.
593 	 *
594 	 * If we need to process the file's pages on I/O, set the file type.
595 	 * If it's a hash file, always call pgin and pgout routines.  This
596 	 * means that hash files can never be mapped into process memory.  If
597 	 * it's a btree file and requires swapping, we need to page the file
598 	 * in and out.  This has to be right -- we can't mmap files that are
599 	 * being paged in and out.
600 	 */
601 	if (type == DB_HASH)
602 		ftype = DB_FTYPE_HASH;
603 	else
604 		ftype = F_ISSET(dbp, DB_AM_SWAP) ? DB_FTYPE_BTREE : 0;
605 	pginfo.db_pagesize = dbp->pgsize;
606 	pginfo.needswap = F_ISSET(dbp, DB_AM_SWAP);
607 	pgcookie.data = &pginfo;
608 	pgcookie.size = sizeof(DB_PGINFO);
609 
610 	/*
611 	 * Set up additional memp_fopen information.
612 	 */
613 	memset(&finfo, 0, sizeof(finfo));
614 	finfo.ftype = ftype;
615 	finfo.pgcookie = &pgcookie;
616 	finfo.fileid = dbp->fileid;
617 	finfo.lsn_offset = 0;
618 	finfo.clear_len = DB_PAGE_CLEAR_LEN;
619 	if ((ret = memp_fopen(dbp->mp, fname,
620 	    F_ISSET(dbp, DB_AM_RDONLY) ? DB_RDONLY : 0,
621 	    0, dbp->pgsize, &finfo, &dbp->mpf)) != 0)
622 		goto err;
623 
624 	/*
625 	 * XXX
626 	 * We need a per-thread mutex that lives in shared memory -- HP-UX
627 	 * can't allocate mutexes in malloc'd memory.  Allocate it from the
628 	 * shared memory region, since it's the only one that is guaranteed
629 	 * to exist.
630 	 */
631 	if (F_ISSET(dbp, DB_AM_THREAD)) {
632 		if ((ret = __memp_reg_alloc(dbp->mp,
633 		    sizeof(db_mutex_t), NULL, &dbp->mutexp)) != 0)
634 			goto err;
635 		/*
636 		 * Since we only get here if DB_THREAD was specified, we know
637 		 * we have spinlocks and no file offset argument is needed.
638 		 */
639 		(void)__db_mutex_init(dbp->mutexp, 0);
640 	}
641 
642 	/* Get a log file id. */
643 	if (F_ISSET(dbp, DB_AM_LOGGING) &&
644 	    (ret = log_register(dbenv->lg_info,
645 	    dbp, fname, type, &dbp->log_fileid)) != 0)
646 		goto err;
647 
648 	/* Call the real open function. */
649 	switch (type) {
650 	case DB_BTREE:
651 		if (dbinfo != NULL && (ret = __db_fchk(dbenv,
652 		    "db_open", dbinfo->flags, DB_RECNUM | DB_DUP)) != 0)
653 			goto err;
654 		if (dbinfo != NULL && (ret = __db_fcchk(dbenv,
655 		    "db_open", dbinfo->flags, DB_DUP, DB_RECNUM)) != 0)
656 			goto err;
657 		if ((ret = __bam_open(dbp, dbinfo)) != 0)
658 			goto err;
659 		break;
660 	case DB_HASH:
661 		if (dbinfo != NULL && (ret = __db_fchk(dbenv,
662 		    "db_open", dbinfo->flags, DB_DUP)) != 0)
663 			goto err;
664 		if ((ret = __ham_open(dbp, dbinfo)) != 0)
665 			goto err;
666 		break;
667 	case DB_RECNO:
668 #define	DB_INFO_FLAGS \
669 	(DB_DELIMITER | DB_FIXEDLEN | DB_PAD | DB_RENUMBER | DB_SNAPSHOT)
670 		if (dbinfo != NULL && (ret = __db_fchk(dbenv,
671 		    "db_open", dbinfo->flags, DB_INFO_FLAGS)) != 0)
672 			goto err;
673 		if ((ret = __ram_open(dbp, dbinfo)) != 0)
674 			goto err;
675 		break;
676 	default:
677 		abort();
678 	}
679 
680 	*dbpp = dbp;
681 	return (0);
682 
683 einval:	ret = EINVAL;
684 err:	/* Close the file descriptor. */
685 	if (fd != -1)
686 		(void)__os_close(fd);
687 
688 	/* Discard the log file id. */
689 	if (dbp->log_fileid != 0)
690 		(void)log_unregister(dbenv->lg_info, dbp->log_fileid);
691 
692 	/* Close the memory pool file. */
693 	if (dbp->mpf != NULL)
694 		(void)memp_fclose(dbp->mpf);
695 
696 	/* If the memory pool was local, close it. */
697 	if (F_ISSET(dbp, DB_AM_MLOCAL) && dbp->mp != NULL)
698 		(void)memp_close(dbp->mp);
699 
700 	/* If we allocated a DB_ENV, discard it. */
701 	if (dbp->mp_dbenv != NULL)
702 		__os_free(dbp->mp_dbenv, sizeof(DB_ENV));
703 
704 	if (real_name != NULL)
705 		__os_freestr(real_name);
706 	if (dbp != NULL)
707 		__os_free(dbp, sizeof(DB));
708 
709 	return (ret);
710 }
711 
712 /*
713  * __db_close --
714  *	Close a DB tree.
715  *
716  * PUBLIC: int __db_close __P((DB *, u_int32_t));
717  */
718 int
__db_close(dbp,flags)719 __db_close(dbp, flags)
720 	DB *dbp;
721 	u_int32_t flags;
722 {
723 	DBC *dbc;
724 	int ret, t_ret;
725 
726 	DB_PANIC_CHECK(dbp);
727 
728 	/* Validate arguments. */
729 	if ((ret = __db_closechk(dbp, flags)) != 0)
730 		return (ret);
731 
732 	/* Sync the underlying file. */
733 	if (flags != DB_NOSYNC &&
734 	    (t_ret = dbp->sync(dbp, 0)) != 0 && ret == 0)
735 		ret = t_ret;
736 
737 	/*
738 	 * Go through the active cursors and call the cursor recycle routine,
739 	 * which resolves pending operations and moves the cursors onto the
740 	 * free list.  Then, walk the free list and call the cursor destroy
741 	 * routine.
742 	 */
743 	while ((dbc = TAILQ_FIRST(&dbp->active_queue)) != NULL)
744 		if ((t_ret = dbc->c_close(dbc)) != 0 && ret == 0)
745 			ret = t_ret;
746 	while ((dbc = TAILQ_FIRST(&dbp->free_queue)) != NULL)
747 		if ((t_ret = __db_c_destroy(dbc)) != 0 && ret == 0)
748 			ret = t_ret;
749 
750 	/* Call the access specific close function. */
751 	if ((t_ret = dbp->am_close(dbp)) != 0 && ret == 0)
752 		ret = t_ret;
753 
754 	/* Sync the memory pool. */
755 	if (flags != DB_NOSYNC && (t_ret = memp_fsync(dbp->mpf)) != 0 &&
756 	    t_ret != DB_INCOMPLETE && ret == 0)
757 		ret = t_ret;
758 
759 	/* Close the memory pool file. */
760 	if ((t_ret = memp_fclose(dbp->mpf)) != 0 && ret == 0)
761 		ret = t_ret;
762 
763 	/* If the memory pool was local, close it. */
764 	if (F_ISSET(dbp, DB_AM_MLOCAL) &&
765 	    (t_ret = memp_close(dbp->mp)) != 0 && ret == 0)
766 		ret = t_ret;
767 
768 	if (dbp->saved_open_fd != -1) {
769 		(void)__os_close(dbp->saved_open_fd);
770 		dbp->saved_open_fd = -1;
771 	}
772 
773 	/* Discard the log file id. */
774 	if (F_ISSET(dbp, DB_AM_LOGGING))
775 		(void)log_unregister(dbp->dbenv->lg_info, dbp->log_fileid);
776 
777 	/* If we allocated a DB_ENV, discard it. */
778 	if (dbp->mp_dbenv != NULL)
779 		__os_free(dbp->mp_dbenv, sizeof(DB_ENV));
780 
781 	/* Free the DB. */
782 	__os_free(dbp, sizeof(*dbp));
783 
784 	return (ret);
785 }
786