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) 1995, 1996
9 *	The President and Fellows of Harvard University.  All rights reserved.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 *    notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 *    notice, this list of conditions and the following disclaimer in the
18 *    documentation and/or other materials provided with the distribution.
19 * 3. All advertising materials mentioning features or use of this software
20 *    must display the following acknowledgement:
21 *	This product includes software developed by the University of
22 *	California, Berkeley and its contributors.
23 * 4. Neither the name of the University nor the names of its contributors
24 *    may be used to endorse or promote products derived from this software
25 *    without specific prior written permission.
26 *
27 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
28 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
29 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
30 * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
31 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
32 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
33 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
34 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
35 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
36 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
37 * SUCH DAMAGE.
38 */
39
40#include "config.h"
41
42#ifndef lint
43static const char sccsid[] = "@(#)log_rec.c	10.26 (Sleepycat) 10/21/98";
44#endif /* not lint */
45
46#ifndef NO_SYSTEM_INCLUDES
47#include <sys/types.h>
48
49#include <errno.h>
50#include <string.h>
51#endif
52
53#include "db_int.h"
54#include "shqueue.h"
55#include "log.h"
56#include "db_dispatch.h"
57#include "common_ext.h"
58
59static int __log_do_open __P((DB_LOG *,
60    u_int8_t *, char *, DBTYPE, u_int32_t));
61static int __log_lid_to_fname __P((DB_LOG *, u_int32_t, FNAME **));
62static int __log_open_file __P((DB_LOG *, __log_register_args *));
63
64/*
65 * PUBLIC: int __log_register_recover
66 * PUBLIC:     __P((DB_LOG *, DBT *, DB_LSN *, int, void *));
67 */
68int
69__log_register_recover(logp, dbtp, lsnp, redo, info)
70	DB_LOG *logp;
71	DBT *dbtp;
72	DB_LSN *lsnp;
73	int redo;
74	void *info;
75{
76	DB_ENTRY *dbe;
77	__log_register_args *argp;
78	int ret;
79
80#ifdef DEBUG_RECOVER
81	__log_register_print(logp, dbtp, lsnp, redo, info);
82#endif
83	COMPQUIET(info, NULL);
84	COMPQUIET(lsnp, NULL);
85
86	F_SET(logp, DBC_RECOVER);
87
88	if ((ret = __log_register_read(dbtp->data, &argp)) != 0)
89		goto out;
90
91	if ((argp->opcode == LOG_CHECKPOINT && redo == TXN_OPENFILES) ||
92	    (argp->opcode == LOG_OPEN &&
93	    (redo == TXN_REDO || redo == TXN_OPENFILES ||
94	     redo == TXN_FORWARD_ROLL)) ||
95	    (argp->opcode == LOG_CLOSE &&
96	    (redo == TXN_UNDO || redo == TXN_BACKWARD_ROLL))) {
97		/*
98		 * If we are redoing an open or undoing a close, then we need
99		 * to open a file.
100		 */
101		ret = __log_open_file(logp, argp);
102		if (ret == ENOENT) {
103			if (redo == TXN_OPENFILES)
104				__db_err(logp->dbenv, "warning: %s: %s",
105				    argp->name.data, strerror(ENOENT));
106			ret = 0;
107		}
108	} else if (argp->opcode != LOG_CHECKPOINT &&
109	    argp->opcode != LOG_CLOSE) {
110		/*
111		 * If we are undoing an open, then we need to close the file.
112		 * Note that we do *not* close the file if we are redoing a
113		 * close, because we do not log the reference counts on log
114		 * files and we may have had the file open multiple times,
115		 * and therefore, this close should just dec a reference
116		 * count.  However, since we only do one open during a
117		 * checkpoint, this will inadvertently close the file.
118		 *
119  		 * If the file is deleted, then we can just ignore this close.
120 		 * Otherwise, we should usually have a valid dbp we should
121  		 * close or whose reference count should be decremented.
122 		 * However, if we shut down without closing a file, we
123		 * may, in fact, not have the file open, and that's OK.
124		 */
125		LOCK_LOGTHREAD(logp);
126		if (argp->id < logp->dbentry_cnt) {
127			dbe = &logp->dbentry[argp->id];
128			if (dbe->dbp != NULL && --dbe->refcount == 0) {
129				ret = dbe->dbp->close(dbe->dbp, 0);
130				if (dbe->name != NULL) {
131					__os_freestr(dbe->name);
132					dbe->name = NULL;
133				}
134				(void)__log_rem_logid(logp, argp->id);
135			}
136		}
137		UNLOCK_LOGTHREAD(logp);
138 	} else if (argp->opcode == LOG_CHECKPOINT && redo == TXN_UNDO &&
139 	    (argp->id >= logp->dbentry_cnt ||
140 	    (!logp->dbentry[argp->id].deleted &&
141 	    logp->dbentry[argp->id].dbp == NULL))) {
142 		/*
143 		 * It's a checkpoint and we are rolling backward.  It
144 		 * is possible that the system was shut down and thus
145 		 * ended with a stable checkpoint; this file was never
146 		 * closed and has therefore not been reopened yet.  If
147 		 * so, we need to try to open it.
148 		 */
149 		ret = __log_open_file(logp, argp);
150 		if (ret == ENOENT) {
151 			__db_err(logp->dbenv, "warning: %s: %s",
152			    argp->name.data, strerror(ENOENT));
153 			ret = 0;
154 		}
155	}
156
157out:	F_CLR(logp, DBC_RECOVER);
158	if (argp != NULL)
159		__os_free(argp, 0);
160	return (ret);
161}
162
163/* Hand coded routines. */
164
165/*
166 * Called during log_register recovery.  Make sure that we have an
167 * entry in the dbentry table for this ndx.
168 * Returns 0 on success, non-zero on error.
169 */
170static int
171__log_open_file(lp, argp)
172	DB_LOG *lp;
173	__log_register_args *argp;
174{
175	DB_ENTRY *dbe;
176
177	if (argp->name.size == 0)
178		return(0);
179
180	/*
181	 * Because of reference counting, we cannot automatically close files
182	 * during recovery, so when we're opening, we have to check that the
183	 * name we are opening is what we expect.  If it's not, then we close
184	 * the old file and open the new one.
185	 */
186	LOCK_LOGTHREAD(lp);
187	if (argp->id < lp->dbentry_cnt)
188		dbe = &lp->dbentry[argp->id];
189	else
190		dbe = NULL;
191
192	if (dbe != NULL && (dbe->deleted == 1 || dbe->dbp != NULL) &&
193	    dbe->name != NULL && argp->name.data != NULL &&
194	    strncmp(argp->name.data, dbe->name, argp->name.size) == 0) {
195
196		dbe->refcount++;
197		UNLOCK_LOGTHREAD(lp);
198		return (0);
199	}
200  	UNLOCK_LOGTHREAD(lp);
201
202	if (dbe != NULL && dbe->dbp != NULL) {
203                (void)dbe->dbp->close(dbe->dbp, 0);
204		if (dbe->name != NULL)
205			__os_freestr(dbe->name);
206                dbe->name = NULL;
207		(void)__log_rem_logid(lp, argp->id);
208        }
209
210
211	return (__log_do_open(lp,
212	    argp->uid.data, argp->name.data, argp->ftype, argp->id));
213}
214
215/*
216 * __log_do_open --
217 * 	Open files referenced in the log.  This is the part of the open that
218 * is not protected by the thread mutex.
219 */
220
221static int
222__log_do_open(lp, uid, name, ftype, ndx)
223	DB_LOG *lp;
224	u_int8_t *uid;
225	char *name;
226	DBTYPE ftype;
227	u_int32_t ndx;
228{
229	DB *dbp;
230	int ret;
231
232	dbp = NULL;
233	if ((ret = db_open(name, ftype, 0, 0, lp->dbenv, NULL, &dbp)) == 0) {
234		/*
235		 * Verify that we are opening the same file that we were
236		 * referring to when we wrote this log record.
237		 */
238		if (memcmp(uid, dbp->fileid, DB_FILE_ID_LEN) != 0) {
239			(void)dbp->close(dbp, 0);
240			dbp = NULL;
241			ret = ENOENT;
242		}
243	}
244
245	if (ret == 0 || ret == ENOENT)
246		(void)__log_add_logid(lp, dbp, name, ndx);
247
248	return (ret);
249}
250
251/*
252 * __log_add_logid --
253 *	Adds a DB entry to the log's DB entry table.
254 *
255 * PUBLIC: int __log_add_logid __P((DB_LOG *, DB *, const char *, u_int32_t));
256 */
257int
258__log_add_logid(logp, dbp, name, ndx)
259	DB_LOG *logp;
260	DB *dbp;
261	const char *name;
262	u_int32_t ndx;
263{
264	u_int32_t i;
265	int ret;
266
267	ret = 0;
268
269	LOCK_LOGTHREAD(logp);
270
271	/*
272	 * Check if we need to grow the table.  Note, ndx is 0-based (the
273	 * index into the DB entry table) an dbentry_cnt is 1-based, the
274	 * number of available slots.
275	 */
276	if (logp->dbentry_cnt <= ndx) {
277		if ((ret = __os_realloc(&logp->dbentry,
278		    (ndx + DB_GROW_SIZE) * sizeof(DB_ENTRY))) != 0)
279			goto err;
280
281		/* Initialize the new entries. */
282		for (i = logp->dbentry_cnt; i < ndx + DB_GROW_SIZE; i++) {
283			logp->dbentry[i].dbp = NULL;
284			logp->dbentry[i].deleted = 0;
285			logp->dbentry[i].name = NULL;
286		}
287
288		logp->dbentry_cnt = i;
289	}
290
291	/* Make space for the name and copy it in. */
292	if (name != NULL) {
293		if ((ret = __os_malloc(strlen(name) + 1,
294		    NULL, &logp->dbentry[ndx].name)) != 0)
295			goto err;
296		strcpy(logp->dbentry[ndx].name, name);
297	}
298
299	if (logp->dbentry[ndx].deleted == 0 && logp->dbentry[ndx].dbp == NULL) {
300		logp->dbentry[ndx].dbp = dbp;
301		logp->dbentry[ndx].refcount = 1;
302		logp->dbentry[ndx].deleted = dbp == NULL;
303	} else
304		logp->dbentry[ndx].refcount++;
305
306
307err:	UNLOCK_LOGTHREAD(logp);
308	return (ret);
309}
310
311
312/*
313 * __db_fileid_to_db --
314 *	Return the DB corresponding to the specified fileid.
315 *
316 * PUBLIC: int __db_fileid_to_db __P((DB_LOG *, DB **, u_int32_t));
317 */
318int
319__db_fileid_to_db(logp, dbpp, ndx)
320	DB_LOG *logp;
321	DB **dbpp;
322	u_int32_t ndx;
323{
324	int ret;
325	char *name;
326	FNAME *fname;
327
328	ret = 0;
329	LOCK_LOGTHREAD(logp);
330
331	/*
332	 * Under XA, a process different than the one issuing DB
333	 * operations may abort a transaction.  In this case,
334	 * recovery routines are run by a process that does not
335	 * necessarily have the file open.  In this case, we must
336	 * open the file explicitly.
337	 */
338	if (ndx >= logp->dbentry_cnt ||
339	    (!logp->dbentry[ndx].deleted && logp->dbentry[ndx].dbp == NULL)) {
340		if (__log_lid_to_fname(logp, ndx, &fname) != 0) {
341			/* Couldn't find entry; this is a fatal error. */
342			ret = EINVAL;
343			goto err;
344		}
345		name = R_ADDR(logp, fname->name_off);
346		/*
347		 * __log_do_open is called without protection of the
348		 * log thread lock.
349		 */
350		UNLOCK_LOGTHREAD(logp);
351		/*
352		 * At this point, we are not holding the thread lock, so
353		 * exit directly instead of going through the exit code
354		 * at the bottom.  If the __log_do_open succeeded, then
355		 * we don't need to do any of the remaining error checking
356		 * at the end of this routine.
357		 */
358		if ((ret = __log_do_open(logp,
359		    fname->ufid, name, fname->s_type, ndx)) != 0)
360			return (ret);
361		*dbpp = logp->dbentry[ndx].dbp;
362		return (0);
363	}
364
365	/*
366	 * Return DB_DELETED if the file has been deleted
367	 * (it's not an error).
368	 */
369	if (logp->dbentry[ndx].deleted) {
370		ret = DB_DELETED;
371		goto err;
372	}
373
374	/*
375	 * Otherwise return 0, but if we don't have a corresponding DB,
376	 * it's an error.
377	 */
378	if ((*dbpp = logp->dbentry[ndx].dbp) == NULL)
379		ret = ENOENT;
380
381err:	UNLOCK_LOGTHREAD(logp);
382	return (ret);
383}
384
385/*
386 * Close files that were opened by the recovery daemon.
387 *
388 * PUBLIC: void __log_close_files __P((DB_LOG *));
389 */
390void
391__log_close_files(logp)
392	DB_LOG *logp;
393{
394	u_int32_t i;
395
396	LOCK_LOGTHREAD(logp);
397	for (i = 0; i < logp->dbentry_cnt; i++)
398		if (logp->dbentry[i].dbp) {
399			logp->dbentry[i].dbp->close(logp->dbentry[i].dbp, 0);
400			logp->dbentry[i].dbp = NULL;
401			logp->dbentry[i].deleted = 0;
402		}
403	F_CLR(logp, DBC_RECOVER);
404	UNLOCK_LOGTHREAD(logp);
405}
406
407/*
408 * PUBLIC: void __log_rem_logid __P((DB_LOG *, u_int32_t));
409 */
410void
411__log_rem_logid(logp, ndx)
412	DB_LOG *logp;
413	u_int32_t ndx;
414{
415	LOCK_LOGTHREAD(logp);
416	if (--logp->dbentry[ndx].refcount == 0) {
417		logp->dbentry[ndx].dbp = NULL;
418		logp->dbentry[ndx].deleted = 0;
419	}
420	UNLOCK_LOGTHREAD(logp);
421}
422
423/*
424 * __log_lid_to_fname --
425 * 	Traverse the shared-memory region looking for the entry that
426 * matches the passed log fileid.  Returns 0 on success; -1 on error.
427 */
428static int
429__log_lid_to_fname(dblp, lid, fnamep)
430	DB_LOG *dblp;
431	u_int32_t lid;
432	FNAME **fnamep;
433{
434	FNAME *fnp;
435
436	for (fnp = SH_TAILQ_FIRST(&dblp->lp->fq, __fname);
437	    fnp != NULL; fnp = SH_TAILQ_NEXT(fnp, q, __fname)) {
438		if (fnp->ref == 0)	/* Entry not in use. */
439			continue;
440		if (fnp->id == lid) {
441			*fnamep = fnp;
442			return (0);
443		}
444	}
445	return (-1);
446}
447