4 * Sleepycat (Berkeley) DB driver for Citadel
8 /*****************************************************************************
9 Tunable configuration parameters for the Sleepycat DB back end
10 *****************************************************************************/
12 /* Citadel will checkpoint the db at the end of every session, but only if
13 * the specified number of kilobytes has been written, or if the specified
14 * number of minutes has passed, since the last checkpoint.
16 #define MAX_CHECKPOINT_KBYTES 256
17 #define MAX_CHECKPOINT_MINUTES 15
19 /*****************************************************************************/
28 #include <sys/types.h>
34 #elif defined(HAVE_DB4_DB_H)
37 #error Neither <db.h> nor <db4/db.h> was found by configure. Install db4-devel.
41 #if DB_VERSION_MAJOR < 4 || DB_VERSION_MINOR < 1
42 #error Citadel requires Berkeley DB v4.1 or newer. Please upgrade.
48 #include "citserver.h"
51 #include "sysdep_decls.h"
55 #include "ctdl_module.h"
58 static DB *dbp[MAXCDB]; /* One DB handle for each Citadel database */
59 static DB_ENV *dbenv; /* The DB environment (global) */
67 /* Verbose logging callback */
68 void cdb_verbose_log(const DB_ENV *dbenv, const char *msg)
70 lprintf(CTDL_DEBUG, "BDB: %s\n", msg);
74 /* Verbose logging callback */
75 void cdb_verbose_err(const DB_ENV *dbenv, const char *errpfx, const char *msg)
77 lprintf(CTDL_ALERT, "BDB: %s\n", msg);
81 /* just a little helper function */
82 static void txabort(DB_TXN * tid)
86 ret = tid->abort(tid);
89 lprintf(CTDL_EMERG, "cdb_*: txn_abort: %s\n",
95 /* this one is even more helpful than the last. */
96 static void txcommit(DB_TXN * tid)
100 ret = tid->commit(tid, 0);
103 lprintf(CTDL_EMERG, "cdb_*: txn_commit: %s\n",
109 /* are you sensing a pattern yet? */
110 static void txbegin(DB_TXN ** tid)
114 ret = dbenv->txn_begin(dbenv, NULL, tid, 0);
117 lprintf(CTDL_EMERG, "cdb_*: txn_begin: %s\n",
123 static void dbpanic(DB_ENV * env, int errval)
125 lprintf(CTDL_EMERG, "cdb_*: Berkeley DB panic: %d\n", errval);
128 static void cclose(DBC * cursor)
132 if ((ret = cursor->c_close(cursor))) {
133 lprintf(CTDL_EMERG, "cdb_*: c_close: %s\n",
139 static void bailIfCursor(DBC ** cursors, const char *msg)
143 for (i = 0; i < MAXCDB; i++)
144 if (cursors[i] != NULL) {
146 "cdb_*: cursor still in progress on cdb %d: %s\n",
152 void check_handles(void *arg)
155 ThreadTSD *tsd = (ThreadTSD *) arg;
157 bailIfCursor(tsd->cursors, "in check_handles");
159 if (tsd->tid != NULL) {
161 "cdb_*: transaction still in progress!");
167 void cdb_check_handles(void)
169 check_handles(pthread_getspecific(ThreadKey));
174 * Reclaim unused space in the databases. We need to do each one of
175 * these discretely, rather than in a loop.
177 * This is a stub function in the Sleepycat DB backend, because there is no
178 * such API call available.
180 void defrag_databases(void)
187 * Cull the database logs
189 static void cdb_cull_logs(void)
198 /* Get the list of names. */
199 if ((ret = dbenv->log_archive(dbenv, &list, flags)) != 0) {
200 lprintf(CTDL_ERR, "cdb_cull_logs: %s\n", db_strerror(ret));
204 /* Print the list of names. */
206 for (file = list; *file != NULL; ++file) {
207 lprintf(CTDL_DEBUG, "Deleting log: %s\n", *file);
210 snprintf(errmsg, sizeof(errmsg),
211 " ** ERROR **\n \n \n "
212 "Citadel was unable to delete the "
213 "database log file '%s' because of the "
214 "following error:\n \n %s\n \n"
215 " This log file is no longer in use "
216 "and may be safely deleted.\n",
217 *file, strerror(errno));
218 aide_message(errmsg, "Database Warning Message");
226 * Manually initiate log file cull.
228 void cmd_cull(char *argbuf) {
229 if (CtdlAccessCheck(ac_internal)) return;
231 cprintf("%d Database log file cull completed.\n", CIT_OK);
236 * Request a checkpoint of the database.
238 void cdb_checkpoint(void)
241 // static time_t last_run = 0L;
243 /* Only do a checkpoint once per minute. */
245 * Don't need this any more, since the thread that calls us sleeps for 60 seconds between calls
247 if ((time(NULL) - last_run) < 60L) {
250 last_run = time(NULL);
253 lprintf(CTDL_DEBUG, "-- db checkpoint --\n");
254 ret = dbenv->txn_checkpoint(dbenv,
255 MAX_CHECKPOINT_KBYTES,
256 MAX_CHECKPOINT_MINUTES, 0);
259 lprintf(CTDL_EMERG, "cdb_checkpoint: txn_checkpoint: %s\n",
264 /* After a successful checkpoint, we can cull the unused logs */
265 if (config.c_auto_cull) {
273 * Open the various databases we'll be using. Any database which
274 * does not exist should be created. Note that we don't need a
275 * critical section here, because there aren't any active threads
276 * manipulating the database yet.
278 void open_databases(void)
282 char dbfilename[SIZ];
285 lprintf(CTDL_DEBUG, "cdb_*: open_databases() starting\n");
286 lprintf(CTDL_DEBUG, "Compiled db: %s\n", DB_VERSION_STRING);
287 lprintf(CTDL_INFO, " Linked db: %s\n",
288 db_version(NULL, NULL, NULL));
290 lprintf(CTDL_INFO, "Linked zlib: %s\n", zlibVersion());
294 * Silently try to create the database subdirectory. If it's
295 * already there, no problem.
297 mkdir(ctdl_data_dir, 0700);
298 chmod(ctdl_data_dir, 0700);
299 chown(ctdl_data_dir, CTDLUID, (-1));
301 lprintf(CTDL_DEBUG, "cdb_*: Setting up DB environment\n");
302 db_env_set_func_yield(sched_yield);
303 ret = db_env_create(&dbenv, 0);
305 lprintf(CTDL_EMERG, "cdb_*: db_env_create: %s\n",
309 dbenv->set_errpfx(dbenv, "citserver");
310 dbenv->set_paniccall(dbenv, dbpanic);
311 dbenv->set_errcall(dbenv, cdb_verbose_err);
312 dbenv->set_errpfx(dbenv, "ctdl");
313 #if (DB_VERSION_MAJOR == 4) && (DB_VERSION_MINOR >= 3)
314 dbenv->set_msgcall(dbenv, cdb_verbose_log);
316 dbenv->set_verbose(dbenv, DB_VERB_DEADLOCK, 1);
317 dbenv->set_verbose(dbenv, DB_VERB_RECOVERY, 1);
320 * We want to specify the shared memory buffer pool cachesize,
321 * but everything else is the default.
323 ret = dbenv->set_cachesize(dbenv, 0, 64 * 1024, 0);
325 lprintf(CTDL_EMERG, "cdb_*: set_cachesize: %s\n",
327 dbenv->close(dbenv, 0);
331 if ((ret = dbenv->set_lk_detect(dbenv, DB_LOCK_DEFAULT))) {
332 lprintf(CTDL_EMERG, "cdb_*: set_lk_detect: %s\n",
334 dbenv->close(dbenv, 0);
338 flags = DB_CREATE | DB_INIT_MPOOL | DB_PRIVATE | DB_INIT_TXN | DB_INIT_LOCK | DB_THREAD | DB_RECOVER;
339 lprintf(CTDL_DEBUG, "dbenv->open(dbenv, %s, %d, 0)\n", ctdl_data_dir, flags);
340 ret = dbenv->open(dbenv, ctdl_data_dir, flags, 0);
341 if (ret == DB_RUNRECOVERY) {
342 lprintf(CTDL_ALERT, "dbenv->open: %s\n", db_strerror(ret));
343 lprintf(CTDL_ALERT, "Attempting recovery...\n");
345 ret = dbenv->open(dbenv, ctdl_data_dir, flags, 0);
347 if (ret == DB_RUNRECOVERY) {
348 lprintf(CTDL_ALERT, "dbenv->open: %s\n", db_strerror(ret));
349 lprintf(CTDL_ALERT, "Attempting catastrophic recovery...\n");
350 flags &= ~DB_RECOVER;
351 flags |= DB_RECOVER_FATAL;
352 ret = dbenv->open(dbenv, ctdl_data_dir, flags, 0);
355 lprintf(CTDL_DEBUG, "dbenv->open: %s\n", db_strerror(ret));
356 dbenv->close(dbenv, 0);
360 lprintf(CTDL_INFO, "Starting up DB\n");
362 for (i = 0; i < MAXCDB; ++i) {
364 /* Create a database handle */
365 ret = db_create(&dbp[i], dbenv, 0);
367 lprintf(CTDL_DEBUG, "db_create: %s\n",
373 /* Arbitrary names for our tables -- we reference them by
374 * number, so we don't have string names for them.
376 snprintf(dbfilename, sizeof dbfilename, "cdb.%02x", i);
378 ret = dbp[i]->open(dbp[i],
383 DB_CREATE | DB_AUTO_COMMIT | DB_THREAD,
386 lprintf(CTDL_EMERG, "db_open[%d]: %s\n", i,
395 /* Make sure we own all the files, because in a few milliseconds
396 * we're going to drop root privs.
398 void cdb_chmod_data(void) {
401 char filename[PATH_MAX];
403 dp = opendir(ctdl_data_dir);
405 while (d = readdir(dp), d != NULL) {
406 if (d->d_name[0] != '.') {
407 snprintf(filename, sizeof filename,
408 "%s/%s", ctdl_data_dir, d->d_name);
409 lprintf(9, "chmod(%s, 0600) returned %d\n",
410 filename, chmod(filename, 0600)
412 lprintf(9, "chown(%s, CTDLUID, -1) returned %d\n",
413 filename, chown(filename, CTDLUID, (-1))
420 lprintf(CTDL_DEBUG, "open_databases() finished\n");
422 CtdlRegisterProtoHook(cmd_cull, "CULL", "Cull database logs");
427 * Close all of the db database files we've opened. This can be done
428 * in a loop, since it's just a bunch of closes.
430 void close_databases(void)
435 ctdl_thread_internal_free_tsd();
437 if ((ret = dbenv->txn_checkpoint(dbenv, 0, 0, 0))) {
439 "txn_checkpoint: %s\n", db_strerror(ret));
442 /* print some statistics... */
444 dbenv->lock_stat_print(dbenv, DB_STAT_ALL);
447 /* close the tables */
448 for (a = 0; a < MAXCDB; ++a) {
449 lprintf(CTDL_INFO, "Closing database %d\n", a);
450 ret = dbp[a]->close(dbp[a], 0);
453 "db_close: %s\n", db_strerror(ret));
458 /* Close the handle. */
459 ret = dbenv->close(dbenv, 0);
462 "DBENV->close: %s\n", db_strerror(ret));
468 * Compression functions only used if we have zlib
472 void cdb_decompress_if_necessary(struct cdbdata *cdb)
474 static int magic = COMPRESS_MAGIC;
475 struct CtdlCompressHeader zheader;
476 char *uncompressed_data;
477 char *compressed_data;
478 uLongf destLen, sourceLen;
482 if (cdb->ptr == NULL)
484 if (memcmp(cdb->ptr, &magic, sizeof(magic)))
487 /* At this point we know we're looking at a compressed item. */
488 memcpy(&zheader, cdb->ptr, sizeof(struct CtdlCompressHeader));
490 compressed_data = cdb->ptr;
491 compressed_data += sizeof(struct CtdlCompressHeader);
493 sourceLen = (uLongf) zheader.compressed_len;
494 destLen = (uLongf) zheader.uncompressed_len;
495 uncompressed_data = malloc(zheader.uncompressed_len);
497 if (uncompress((Bytef *) uncompressed_data,
498 (uLongf *) & destLen,
499 (const Bytef *) compressed_data,
500 (uLong) sourceLen) != Z_OK) {
501 lprintf(CTDL_EMERG, "uncompress() error\n");
506 cdb->len = (size_t) destLen;
507 cdb->ptr = uncompressed_data;
510 #endif /* HAVE_ZLIB */
514 * Store a piece of data. Returns 0 if the operation was successful. If a
515 * key already exists it should be overwritten.
517 int cdb_store(int cdb, void *ckey, int ckeylen, void *cdata, int cdatalen)
525 struct CtdlCompressHeader zheader;
526 char *compressed_data = NULL;
528 size_t buffer_len = 0;
532 memset(&dkey, 0, sizeof(DBT));
533 memset(&ddata, 0, sizeof(DBT));
536 ddata.size = cdatalen;
540 /* Only compress Visit records. Everything else is uncompressed. */
541 if (cdb == CDB_VISIT) {
543 zheader.magic = COMPRESS_MAGIC;
544 zheader.uncompressed_len = cdatalen;
545 buffer_len = ((cdatalen * 101) / 100) + 100
546 + sizeof(struct CtdlCompressHeader);
547 destLen = (uLongf) buffer_len;
548 compressed_data = malloc(buffer_len);
549 if (compress2((Bytef *) (compressed_data +
551 CtdlCompressHeader)),
552 &destLen, (Bytef *) cdata, (uLongf) cdatalen,
554 lprintf(CTDL_EMERG, "compress2() error\n");
557 zheader.compressed_len = (size_t) destLen;
558 memcpy(compressed_data, &zheader,
559 sizeof(struct CtdlCompressHeader));
560 ddata.size = (size_t) (sizeof(struct CtdlCompressHeader) +
561 zheader.compressed_len);
562 ddata.data = compressed_data;
567 ret = dbp[cdb]->put(dbp[cdb], /* db */
568 MYTID, /* transaction ID */
573 lprintf(CTDL_EMERG, "cdb_store(%d): %s\n", cdb,
579 free(compressed_data);
584 bailIfCursor(MYCURSORS,
585 "attempt to write during r/o cursor");
590 if ((ret = dbp[cdb]->put(dbp[cdb], /* db */
591 tid, /* transaction ID */
595 if (ret == DB_LOCK_DEADLOCK) {
599 lprintf(CTDL_EMERG, "cdb_store(%d): %s\n",
600 cdb, db_strerror(ret));
607 free(compressed_data);
616 * Delete a piece of data. Returns 0 if the operation was successful.
618 int cdb_delete(int cdb, void *key, int keylen)
625 memset(&dkey, 0, sizeof dkey);
630 ret = dbp[cdb]->del(dbp[cdb], MYTID, &dkey, 0);
632 lprintf(CTDL_EMERG, "cdb_delete(%d): %s\n", cdb,
634 if (ret != DB_NOTFOUND)
638 bailIfCursor(MYCURSORS,
639 "attempt to delete during r/o cursor");
644 if ((ret = dbp[cdb]->del(dbp[cdb], tid, &dkey, 0))
645 && ret != DB_NOTFOUND) {
646 if (ret == DB_LOCK_DEADLOCK) {
650 lprintf(CTDL_EMERG, "cdb_delete(%d): %s\n",
651 cdb, db_strerror(ret));
661 static DBC *localcursor(int cdb)
666 if (MYCURSORS[cdb] == NULL)
667 ret = dbp[cdb]->cursor(dbp[cdb], MYTID, &curs, 0);
670 MYCURSORS[cdb]->c_dup(MYCURSORS[cdb], &curs,
674 lprintf(CTDL_EMERG, "localcursor: %s\n", db_strerror(ret));
683 * Fetch a piece of data. If not found, returns NULL. Otherwise, it returns
684 * a struct cdbdata which it is the caller's responsibility to free later on
685 * using the cdb_free() routine.
687 struct cdbdata *cdb_fetch(int cdb, void *key, int keylen)
690 struct cdbdata *tempcdb;
694 memset(&dkey, 0, sizeof(DBT));
699 memset(&dret, 0, sizeof(DBT));
700 dret.flags = DB_DBT_MALLOC;
701 ret = dbp[cdb]->get(dbp[cdb], MYTID, &dkey, &dret, 0);
706 memset(&dret, 0, sizeof(DBT));
707 dret.flags = DB_DBT_MALLOC;
709 curs = localcursor(cdb);
711 ret = curs->c_get(curs, &dkey, &dret, DB_SET);
714 while (ret == DB_LOCK_DEADLOCK);
718 if ((ret != 0) && (ret != DB_NOTFOUND)) {
719 lprintf(CTDL_EMERG, "cdb_fetch(%d): %s\n", cdb,
726 tempcdb = (struct cdbdata *) malloc(sizeof(struct cdbdata));
728 if (tempcdb == NULL) {
730 "cdb_fetch: Cannot allocate memory for tempcdb\n");
734 tempcdb->len = dret.size;
735 tempcdb->ptr = dret.data;
737 cdb_decompress_if_necessary(tempcdb);
744 * Free a cdbdata item.
746 * Note that we only free the 'ptr' portion if it is not NULL. This allows
747 * other code to assume ownership of that memory simply by storing the
748 * pointer elsewhere and then setting 'ptr' to NULL. cdb_free() will then
751 void cdb_free(struct cdbdata *cdb)
759 void cdb_close_cursor(int cdb)
761 if (MYCURSORS[cdb] != NULL)
762 cclose(MYCURSORS[cdb]);
764 MYCURSORS[cdb] = NULL;
768 * Prepare for a sequential search of an entire database.
769 * (There is guaranteed to be no more than one traversal in
770 * progress per thread at any given time.)
772 void cdb_rewind(int cdb)
776 if (MYCURSORS[cdb] != NULL) {
778 "cdb_rewind: must close cursor on database %d before reopening.\n",
781 /* cclose(MYCURSORS[cdb]); */
785 * Now initialize the cursor
787 ret = dbp[cdb]->cursor(dbp[cdb], MYTID, &MYCURSORS[cdb], 0);
789 lprintf(CTDL_EMERG, "cdb_rewind: db_cursor: %s\n",
797 * Fetch the next item in a sequential search. Returns a pointer to a
798 * cdbdata structure, or NULL if we've hit the end.
800 struct cdbdata *cdb_next_item(int cdb)
803 struct cdbdata *cdbret;
806 /* Initialize the key/data pair so the flags aren't set. */
807 memset(&key, 0, sizeof(key));
808 memset(&data, 0, sizeof(data));
809 data.flags = DB_DBT_MALLOC;
811 ret = MYCURSORS[cdb]->c_get(MYCURSORS[cdb], &key, &data, DB_NEXT);
814 if (ret != DB_NOTFOUND) {
815 lprintf(CTDL_EMERG, "cdb_next_item(%d): %s\n",
816 cdb, db_strerror(ret));
819 cclose(MYCURSORS[cdb]);
820 MYCURSORS[cdb] = NULL;
821 return NULL; /* presumably, end of file */
824 cdbret = (struct cdbdata *) malloc(sizeof(struct cdbdata));
825 cdbret->len = data.size;
826 cdbret->ptr = data.data;
828 cdb_decompress_if_necessary(cdbret);
837 * Transaction-based stuff. I'm writing this as I bake cookies...
840 void cdb_begin_transaction(void)
843 bailIfCursor(MYCURSORS,
844 "can't begin transaction during r/o cursor");
848 "cdb_begin_transaction: ERROR: nested transaction\n");
855 void cdb_end_transaction(void)
859 for (i = 0; i < MAXCDB; i++)
860 if (MYCURSORS[i] != NULL) {
861 lprintf(CTDL_WARNING,
862 "cdb_end_transaction: WARNING: cursor %d still open at transaction end\n",
864 cclose(MYCURSORS[i]);
870 "cdb_end_transaction: ERROR: txcommit(NULL) !!\n");
879 * Truncate (delete every record)
881 void cdb_trunc(int cdb)
889 "cdb_trunc must not be called in a transaction.\n");
892 bailIfCursor(MYCURSORS,
893 "attempt to write during r/o cursor");
898 if ((ret = dbp[cdb]->truncate(dbp[cdb], /* db */
899 NULL, /* transaction ID */
900 &count, /* #rows deleted */
902 if (ret == DB_LOCK_DEADLOCK) {
907 "cdb_truncate(%d): %s\n", cdb,