1 /*-
2 * See the file LICENSE for redistribution information.
3 *
4 * Copyright (c) 1996, 1997, 1998, 1999
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. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 */
35
36 #include "db_config.h"
37
38 #ifndef lint
39 static const char sccsid[] = "@(#)log_rec.c 11.16 (Sleepycat) 10/19/99";
40 #endif /* not lint */
41
42 #ifndef NO_SYSTEM_INCLUDES
43 #include <sys/types.h>
44
45 #include <assert.h>
46 #include <errno.h>
47 #include <string.h>
48 #endif
49
50 #include "db_int.h"
51 #include "log.h"
52 #include "db_dispatch.h"
53 #include "db_page.h"
54 #include "db_ext.h"
55
56 static int CDB___log_do_open __P((DB_LOG *, u_int8_t *, char *, DBTYPE, u_int32_t));
57 static int CDB___log_lid_to_fname __P((DB_LOG *, int32_t, FNAME **));
58 static int CDB___log_open_file __P((DB_LOG *, __log_register_args *));
59
60 /*
61 * PUBLIC: int CDB___log_register_recover
62 * PUBLIC: __P((DB_ENV *, DBT *, DB_LSN *, int, void *));
63 */
64 int
CDB___log_register_recover(dbenv,dbtp,lsnp,redo,info)65 CDB___log_register_recover(dbenv, dbtp, lsnp, redo, info)
66 DB_ENV *dbenv;
67 DBT *dbtp;
68 DB_LSN *lsnp;
69 int redo;
70 void *info;
71 {
72 DB_ENTRY *dbe;
73 DB_LOG *logp;
74 __log_register_args *argp;
75 int do_rem, ret, t_ret;
76
77 logp = dbenv->lg_handle;
78
79 #ifdef DEBUG_RECOVER
80 CDB___log_register_print(logp, dbtp, lsnp, redo, info);
81 #endif
82 COMPQUIET(lsnp, NULL);
83
84 F_SET(logp, DBC_RECOVER);
85
86 if ((ret = CDB___log_register_read(dbtp->data, &argp)) != 0)
87 goto out;
88
89 if ((argp->opcode == LOG_OPEN &&
90 (redo == TXN_REDO || redo == TXN_OPENFILES ||
91 redo == TXN_FORWARD_ROLL)) ||
92 (argp->opcode == LOG_CLOSE &&
93 (redo == TXN_UNDO || redo == TXN_BACKWARD_ROLL))) {
94 /*
95 * If we are redoing an open or undoing a close, then we need
96 * to open a file.
97 */
98 ret = CDB___log_open_file(logp, argp);
99 if (ret == ENOENT || ret == EINVAL) {
100 if (redo == TXN_OPENFILES && argp->name.size != 0 &&
101 (ret = CDB___db_txnlist_delete(info,
102 argp->name.data, argp->id, 0)) != 0)
103 goto out;
104 ret = 0;
105 }
106 } else if (argp->opcode != LOG_CHECKPOINT) {
107 /*
108 * If we are undoing an open, then we need to close the file.
109 *
110 * If the file is deleted, then we can just ignore this close.
111 * Otherwise, we should usually have a valid dbp we should
112 * close or whose reference count should be decremented.
113 * However, if we shut down without closing a file, we may, in
114 * fact, not have the file open, and that's OK.
115 */
116 do_rem = 0;
117 MUTEX_THREAD_LOCK(logp->mutexp);
118 if (argp->id < logp->dbentry_cnt) {
119 dbe = &logp->dbentry[argp->id];
120 #ifdef DIAGNOSTIC
121 assert(dbe->refcount == 1);
122 #endif
123 ret = CDB___db_txnlist_close(info, argp->id, dbe->count);
124 if (dbe->dbp != NULL &&
125 (t_ret = dbe->dbp->close(dbe->dbp, 0)) != 0
126 && ret == 0)
127 ret = t_ret;
128 do_rem = 1;
129 }
130 MUTEX_THREAD_UNLOCK(logp->mutexp);
131 if (do_rem)
132 (void)CDB___log_rem_logid(logp, argp->id);
133 } else if ((redo == TXN_UNDO || redo == TXN_OPENFILES) &&
134 (argp->id >= logp->dbentry_cnt ||
135 (!logp->dbentry[argp->id].deleted &&
136 logp->dbentry[argp->id].dbp == NULL))) {
137 /*
138 * It's a checkpoint and we are rolling backward. It
139 * is possible that the system was shut down and thus
140 * ended with a stable checkpoint; this file was never
141 * closed and has therefore not been reopened yet. If
142 * so, we need to try to open it.
143 */
144 ret = CDB___log_open_file(logp, argp);
145 if (ret == ENOENT || ret == EINVAL) {
146 if (argp->name.size != 0 && (ret =
147 CDB___db_txnlist_delete(info,
148 argp->name.data, argp->id, 0)) != 0)
149 goto out;
150 ret = 0;
151 }
152 }
153
154 out: F_CLR(logp, DBC_RECOVER);
155 if (argp != NULL)
156 CDB___os_free(argp, 0);
157 return (ret);
158 }
159
160 /*
161 * CDB___log_open_file --
162 * Called during CDB_log_register recovery. Make sure that we have an
163 * entry in the dbentry table for this ndx. Returns 0 on success,
164 * non-zero on error.
165 */
166 static int
CDB___log_open_file(lp,argp)167 CDB___log_open_file(lp, argp)
168 DB_LOG *lp;
169 __log_register_args *argp;
170 {
171 DB_ENTRY *dbe;
172
173 /*
174 * We never re-open temporary files. Temp files are only
175 * useful during aborts in which case the dbp was entered
176 * when the file was registered. During recovery, we treat
177 * temp files as properly deleted files, allowing the open to
178 * fail and not reporting any errors when recovery fails to
179 * get a valid dbp from db_fileid_to_db.
180 */
181 if (argp->name.size == 0) {
182 (void)CDB___log_add_logid(lp, NULL, argp->id);
183 return (ENOENT);
184 }
185
186 /*
187 * Because of reference counting, we cannot automatically close files
188 * during recovery, so when we're opening, we have to check that the
189 * name we are opening is what we expect. If it's not, then we close
190 * the old file and open the new one.
191 */
192 MUTEX_THREAD_LOCK(lp->mutexp);
193 if (argp->id < lp->dbentry_cnt)
194 dbe = &lp->dbentry[argp->id];
195 else
196 dbe = NULL;
197
198 if (dbe != NULL && (dbe->deleted == 1 || dbe->dbp != NULL)) {
199 dbe->refcount++;
200 MUTEX_THREAD_UNLOCK(lp->mutexp);
201 return (0);
202 }
203
204 MUTEX_THREAD_UNLOCK(lp->mutexp);
205
206 return (CDB___log_do_open(lp,
207 argp->uid.data, argp->name.data, argp->ftype, argp->id));
208 }
209
210 /*
211 * CDB___log_do_open --
212 * Open files referenced in the log. This is the part of the open that
213 * is not protected by the thread mutex.
214 */
215 static int
CDB___log_do_open(lp,uid,name,ftype,ndx)216 CDB___log_do_open(lp, uid, name, ftype, ndx)
217 DB_LOG *lp;
218 u_int8_t *uid;
219 char *name;
220 DBTYPE ftype;
221 u_int32_t ndx;
222 {
223 DB *dbp;
224 int ret;
225 u_int8_t zeroid[DB_FILE_ID_LEN];
226
227 if ((ret = CDB_db_create(&dbp, lp->dbenv, 0)) != 0)
228 return (ret);
229 if ((ret = dbp->open(dbp, name, NULL, ftype, 0, 0600)) == 0) {
230 /*
231 * Verify that we are opening the same file that we were
232 * referring to when we wrote this log record.
233 */
234 memset(zeroid, 0, DB_FILE_ID_LEN);
235 if (memcmp(uid, dbp->fileid, DB_FILE_ID_LEN) == 0 ||
236 memcmp(dbp->fileid, zeroid, DB_FILE_ID_LEN) == 0) {
237 (void)CDB___log_add_logid(lp, dbp, ndx);
238 return (0);
239 }
240 }
241 (void)dbp->close(dbp, 0);
242 (void)CDB___log_add_logid(lp, NULL, ndx);
243
244 return (ENOENT);
245 }
246
247 /*
248 * CDB___log_add_logid --
249 * Adds a DB entry to the log's DB entry table.
250 *
251 * PUBLIC: int CDB___log_add_logid __P((DB_LOG *, DB *, u_int32_t));
252 */
253 int
CDB___log_add_logid(logp,dbp,ndx)254 CDB___log_add_logid(logp, dbp, ndx)
255 DB_LOG *logp;
256 DB *dbp;
257 u_int32_t ndx;
258 {
259 u_int32_t i;
260 int ret;
261
262 ret = 0;
263
264 MUTEX_THREAD_LOCK(logp->mutexp);
265
266 /*
267 * Check if we need to grow the table. Note, ndx is 0-based (the
268 * index into the DB entry table) an dbentry_cnt is 1-based, the
269 * number of available slots.
270 */
271 if (logp->dbentry_cnt <= ndx) {
272 if ((ret = CDB___os_realloc((ndx + DB_GROW_SIZE) * sizeof(DB_ENTRY),
273 NULL, &logp->dbentry)) != 0)
274 goto err;
275
276 /* Initialize the new entries. */
277 for (i = logp->dbentry_cnt; i < ndx + DB_GROW_SIZE; i++) {
278 logp->dbentry[i].count = 0;
279 logp->dbentry[i].dbp = NULL;
280 logp->dbentry[i].deleted = 0;
281 logp->dbentry[i].refcount = 0;
282 }
283
284 logp->dbentry_cnt = i;
285 }
286
287 if (logp->dbentry[ndx].deleted == 0 &&
288 logp->dbentry[ndx].dbp == NULL) {
289 logp->dbentry[ndx].count = 0;
290 logp->dbentry[ndx].dbp = dbp;
291 logp->dbentry[ndx].deleted = dbp == NULL;
292 logp->dbentry[ndx].refcount = 1;
293 } else
294 logp->dbentry[ndx].refcount++;
295
296
297 err: MUTEX_THREAD_UNLOCK(logp->mutexp);
298 return (ret);
299 }
300
301 /*
302 * CDB___db_fileid_to_db --
303 * Return the DB corresponding to the specified fileid.
304 *
305 * PUBLIC: int CDB___db_fileid_to_db __P((DB_ENV *, DB **, int32_t, int));
306 */
307 int
CDB___db_fileid_to_db(dbenv,dbpp,ndx,inc)308 CDB___db_fileid_to_db(dbenv, dbpp, ndx, inc)
309 DB_ENV *dbenv;
310 DB **dbpp;
311 int32_t ndx;
312 int inc;
313 {
314 DB_LOG *logp;
315 FNAME *fname;
316 int ret;
317 char *name;
318
319 ret = 0;
320 logp = dbenv->lg_handle;
321
322 MUTEX_THREAD_LOCK(logp->mutexp);
323
324 /*
325 * Under XA, a process different than the one issuing DB operations
326 * may abort a transaction. In this case, recovery routines are run
327 * by a process that does not necessarily have the file open, so we
328 * we must open the file explicitly.
329 */
330 if ((u_int32_t)ndx >= logp->dbentry_cnt ||
331 (!logp->dbentry[ndx].deleted && logp->dbentry[ndx].dbp == NULL)) {
332 if (CDB___log_lid_to_fname(logp, ndx, &fname) != 0) {
333 /* Couldn't find entry; this is a fatal error. */
334 ret = EINVAL;
335 goto err;
336 }
337 name = R_ADDR(&logp->reginfo, fname->name_off);
338
339 /*
340 * CDB___log_do_open is called without protection of the
341 * log thread lock.
342 */
343 MUTEX_THREAD_UNLOCK(logp->mutexp);
344
345 /*
346 * At this point, we are not holding the thread lock, so exit
347 * directly instead of going through the exit code at the
348 * bottom. If the CDB___log_do_open succeeded, then we don't need
349 * to do any of the remaining error checking at the end of this
350 * routine.
351 */
352 if ((ret = CDB___log_do_open(logp,
353 fname->ufid, name, fname->s_type, ndx)) != 0)
354 return (ret);
355
356 *dbpp = logp->dbentry[ndx].dbp;
357 return (0);
358 }
359
360 /*
361 * Return DB_DELETED if the file has been deleted (it's not an error).
362 */
363 if (logp->dbentry[ndx].deleted) {
364 ret = DB_DELETED;
365 if (inc)
366 logp->dbentry[ndx].count++;
367 goto err;
368 }
369
370 /*
371 * Otherwise return 0, but if we don't have a corresponding DB, it's
372 * an error.
373 */
374 if ((*dbpp = logp->dbentry[ndx].dbp) == NULL)
375 ret = ENOENT;
376
377 err: MUTEX_THREAD_UNLOCK(logp->mutexp);
378 return (ret);
379 }
380
381 /*
382 * Close files that were opened by the recovery daemon.
383 *
384 * PUBLIC: void CDB___log_close_files __P((DB_ENV *));
385 */
386 void
CDB___log_close_files(dbenv)387 CDB___log_close_files(dbenv)
388 DB_ENV *dbenv;
389 {
390 DB_ENTRY *dbe;
391 DB_LOG *logp;
392 u_int32_t i;
393
394 logp = dbenv->lg_handle;
395 MUTEX_THREAD_LOCK(logp->mutexp);
396 F_SET(logp, DBC_RECOVER);
397 for (i = 0; i < logp->dbentry_cnt; i++) {
398 dbe = &logp->dbentry[i];
399 if (dbe->dbp != NULL) {
400 (void)dbe->dbp->close(dbe->dbp, 0);
401 dbe->dbp = NULL;
402 }
403 dbe->deleted = 0;
404 dbe->refcount = 0;
405 }
406 F_CLR(logp, DBC_RECOVER);
407 MUTEX_THREAD_UNLOCK(logp->mutexp);
408 }
409
410 /*
411 * PUBLIC: void CDB___log_rem_logid __P((DB_LOG *, u_int32_t));
412 */
413 void
CDB___log_rem_logid(logp,ndx)414 CDB___log_rem_logid(logp, ndx)
415 DB_LOG *logp;
416 u_int32_t ndx;
417 {
418 MUTEX_THREAD_LOCK(logp->mutexp);
419 if (--logp->dbentry[ndx].refcount == 0) {
420 logp->dbentry[ndx].dbp = NULL;
421 logp->dbentry[ndx].deleted = 0;
422 }
423 MUTEX_THREAD_UNLOCK(logp->mutexp);
424 }
425
426 /*
427 * CDB___log_lid_to_fname --
428 * Traverse the shared-memory region looking for the entry that
429 * matches the passed log fileid. Returns 0 on success; -1 on error.
430 */
431 static int
CDB___log_lid_to_fname(dblp,lid,fnamep)432 CDB___log_lid_to_fname(dblp, lid, fnamep)
433 DB_LOG *dblp;
434 int32_t lid;
435 FNAME **fnamep;
436 {
437 FNAME *fnp;
438 LOG *lp;
439
440 lp = dblp->reginfo.primary;
441
442 for (fnp = SH_TAILQ_FIRST(&lp->fq, __fname);
443 fnp != NULL; fnp = SH_TAILQ_NEXT(fnp, q, __fname)) {
444 if (fnp->ref == 0) /* Entry not in use. */
445 continue;
446 if (fnp->id == lid) {
447 *fnamep = fnp;
448 return (0);
449 }
450 }
451 return (-1);
452 }
453