1 // This is a data store backend for the Citadel server which uses Berkeley DB.
3 // Copyright (c) 1987-2023 by the citadel.org team
5 // This program is open source software. Use, duplication, or disclosure
6 // is subject to the terms of the GNU General Public License, version 3.
8 // Citadel will checkpoint the db at the end of every session, but only if
9 // the specified number of kilobytes has been written, or if the specified
10 // number of minutes has passed, since the last checkpoint.
11 #define MAX_CHECKPOINT_KBYTES 256
12 #define MAX_CHECKPOINT_MINUTES 15
14 #include "../../sysdep.h"
19 #include <libcitadel.h>
20 #include "../../citserver.h"
21 #include "../../config.h"
22 #include "berkeley_db.h"
24 #if DB_VERSION_MAJOR < 18
25 #error Citadel requires Berkeley DB v18.0 or newer. Please upgrade.
28 // Globals (these are used across multiple functions in *this* module, but not elsewhere)
29 static DB *dbp[MAXCDB]; // One DB handle for each Citadel database
30 static DB_ENV *dbenv; // The DB environment (global)
33 // Thread-Specific-Storage items for this backend
35 DB_TXN *tid; // Transaction handle
36 DBC *cursors[MAXCDB]; // Cursors, for traversals...
40 pthread_key_t bdb_thread_key;
41 #define TSD bdb_get_tsd()
44 // Return a pointer to our thread-specific (not session-specific) data.
45 struct bdb_tsd *bdb_get_tsd(void) {
47 struct bdb_tsd *c = (struct bdb_tsd *) pthread_getspecific(bdb_thread_key) ;
52 // If there's no TSD for this thread, it must be a new thread. Create our TSD region.
53 c = (struct bdb_tsd *) malloc(sizeof(struct bdb_tsd));
54 memset(c, 0, sizeof(struct bdb_tsd));
55 pthread_setspecific(bdb_thread_key, (const void *) c);
60 // Called by other functions in this module to GTFO quickly if we need to. Not part of the backend API.
61 void bdb_abort(void) {
62 syslog(LOG_DEBUG, "bdb: citserver is stopping in order to prevent data loss. uid=%d gid=%d euid=%d egid=%d",
63 getuid(), getgid(), geteuid(), getegid()
65 raise(SIGABRT); // This will exit in a way that can produce a core dump if needed.
66 exit(CTDLEXIT_DB); // Exit if the signal failed to end the program.
70 // Verbose logging callback for Berkeley DB. Not part of the backend API.
71 void bdb_verbose_log(const DB_ENV *dbenv, const char *msg, const char *foo) {
72 if (!IsEmptyStr(msg)) {
73 syslog(LOG_DEBUG, "bdb: %s %s", msg, foo);
78 // Verbose error logging callback for Berkeley DB. Not part of the backend API.
79 void bdb_verbose_err(const DB_ENV *dbenv, const char *errpfx, const char *msg) {
80 syslog(LOG_ERR, "bdb: %s", msg);
84 // Wrapper for txn_abort() that logs/aborts on error. Not part of the backend API.
85 static void bdb_txabort(DB_TXN *tid) {
88 ret = tid->abort(tid);
91 syslog(LOG_ERR, "bdb: txn_abort: %s", db_strerror(ret));
97 // Wrapper for txn_commit() that logs/aborts on error. Not part of the backend API.
98 static void bdb_txcommit(DB_TXN *tid) {
101 ret = tid->commit(tid, 0);
104 syslog(LOG_ERR, "bdb: txn_commit: %s", db_strerror(ret));
110 // Wrapper for txn_begin() that logs/aborts on error. Not part of the backend API.
111 static void bdb_txbegin(DB_TXN **tid) {
114 ret = dbenv->txn_begin(dbenv, NULL, tid, 0);
117 syslog(LOG_ERR, "bdb: txn_begin: %s", db_strerror(ret));
123 // Panic callback for Berkeley DB. Not part of the backend API.
124 static void bdb_dbpanic(DB_ENV *env, int errval) {
125 syslog(LOG_ERR, "bdb: PANIC: %s", db_strerror(errval));
130 // Close a cursor -- not part of the backend API.
131 static void bdb_cclose(DBC *cursor) {
134 if ((ret = cursor->c_close(cursor))) {
135 syslog(LOG_ERR, "bdb: c_close: %s", db_strerror(ret));
141 // Convenience function to abort if a cursor is still open when we didn't expect it to be.
142 // This is not part of the backend API.
143 static void bdb_bailIfCursor(DBC **cursors, const char *msg) {
146 for (i = 0; i < MAXCDB; i++)
147 if (cursors[i] != NULL) {
148 syslog(LOG_ERR, "bdb: cursor still in progress on cdb %02x: %s", i, msg);
154 void bdb_check_handles(void) {
155 bdb_bailIfCursor(TSD->cursors, "in check_handles");
157 if (TSD->tid != NULL) {
158 syslog(LOG_ERR, "bdb: transaction still in progress!");
164 // Request a checkpoint of the database. Called once per minute by the thread manager.
165 void bdb_checkpoint(void) {
168 syslog(LOG_DEBUG, "bdb: -- checkpoint --");
169 ret = dbenv->txn_checkpoint(dbenv, MAX_CHECKPOINT_KBYTES, MAX_CHECKPOINT_MINUTES, 0);
172 syslog(LOG_ERR, "bdb: bdb_checkpoint() txn_checkpoint: %s", db_strerror(ret));
176 // After a successful checkpoint, we can cull the unused logs
177 if (CtdlGetConfigInt("c_auto_cull")) {
178 ret = dbenv->log_set_config(dbenv, DB_LOG_AUTO_REMOVE, 1);
181 ret = dbenv->log_set_config(dbenv, DB_LOG_AUTO_REMOVE, 0);
186 // Open the various databases we'll be using. Any database which
187 // does not exist should be created. Note that we don't need a
188 // critical section here, because there aren't any active threads
189 // manipulating the database yet.
190 void bdb_open_databases(void) {
195 int dbversion_major, dbversion_minor, dbversion_patch;
197 syslog(LOG_DEBUG, "bdb: bdb_open_databases() starting");
198 syslog(LOG_DEBUG, "bdb: Linked zlib: %s", zlibVersion());
199 syslog(LOG_DEBUG, "bdb: Compiled libdb: %s", DB_VERSION_STRING);
200 syslog(LOG_DEBUG, "bdb: Linked libdb: %s", db_version(&dbversion_major, &dbversion_minor, &dbversion_patch));
202 // Create synthetic integer version numbers and compare them.
203 // Never allow citserver to run with a libdb older then the one with which it was compiled.
204 int compiled_db_version = ( (DB_VERSION_MAJOR * 1000000) + (DB_VERSION_MINOR * 1000) + (DB_VERSION_PATCH) );
205 int linked_db_version = ( (dbversion_major * 1000000) + (dbversion_minor * 1000) + (dbversion_patch) );
206 if (compiled_db_version > linked_db_version) {
207 syslog(LOG_ERR, "bdb: citserver is running with a version of libdb older than the one with which it was compiled.");
208 syslog(LOG_ERR, "bdb: This is an invalid configuration. citserver will now exit to prevent data loss.");
212 syslog(LOG_DEBUG, "bdb: Setting up DB environment");
213 ret = db_env_create(&dbenv, 0);
215 syslog(LOG_ERR, "bdb: db_env_create: %s", db_strerror(ret));
216 syslog(LOG_ERR, "bdb: exit code %d", ret);
219 dbenv->set_errpfx(dbenv, "citserver");
220 dbenv->set_paniccall(dbenv, bdb_dbpanic);
221 dbenv->set_errcall(dbenv, bdb_verbose_err);
222 dbenv->set_msgcall(dbenv, bdb_verbose_log);
223 dbenv->set_verbose(dbenv, DB_VERB_DEADLOCK, 1);
224 dbenv->set_verbose(dbenv, DB_VERB_RECOVERY, 1);
226 // We want to specify the shared memory buffer pool cachesize, but everything else is the default.
227 ret = dbenv->set_cachesize(dbenv, 0, 64 * 1024, 0);
229 syslog(LOG_ERR, "bdb: set_cachesize: %s", db_strerror(ret));
230 dbenv->close(dbenv, 0);
231 syslog(LOG_ERR, "bdb: exit code %d", ret);
235 if ((ret = dbenv->set_lk_detect(dbenv, DB_LOCK_DEFAULT))) {
236 syslog(LOG_ERR, "bdb: set_lk_detect: %s", db_strerror(ret));
237 dbenv->close(dbenv, 0);
238 syslog(LOG_ERR, "bdb: exit code %d", ret);
242 flags = DB_CREATE | DB_INIT_MPOOL | DB_PRIVATE | DB_INIT_TXN | DB_INIT_LOCK | DB_THREAD | DB_INIT_LOG;
243 syslog(LOG_DEBUG, "bdb: dbenv->open(dbenv, %s, %d, 0)", ctdl_db_dir, flags);
244 ret = dbenv->open(dbenv, ctdl_db_dir, flags, 0); // try opening the database cleanly
245 if (ret == DB_RUNRECOVERY) {
246 syslog(LOG_ERR, "bdb: dbenv->open: %s", db_strerror(ret));
247 syslog(LOG_ERR, "bdb: attempting recovery...");
249 ret = dbenv->open(dbenv, ctdl_db_dir, flags, 0); // try recovery
251 if (ret == DB_RUNRECOVERY) {
252 syslog(LOG_ERR, "bdb: dbenv->open: %s", db_strerror(ret));
253 syslog(LOG_ERR, "bdb: attempting catastrophic recovery...");
254 flags &= ~DB_RECOVER;
255 flags |= DB_RECOVER_FATAL;
256 ret = dbenv->open(dbenv, ctdl_db_dir, flags, 0); // try catastrophic recovery
259 syslog(LOG_ERR, "bdb: dbenv->open: %s", db_strerror(ret));
260 dbenv->close(dbenv, 0);
261 syslog(LOG_ERR, "bdb: exit code %d", ret);
265 syslog(LOG_INFO, "bdb: mounting databases");
266 for (i = 0; i < MAXCDB; ++i) {
267 ret = db_create(&dbp[i], dbenv, 0); // Create a database handle
269 syslog(LOG_ERR, "bdb: db_create: %s", db_strerror(ret));
270 syslog(LOG_ERR, "bdb: exit code %d", ret);
274 snprintf(dbfilename, sizeof dbfilename, "cdb.%02x", i); // table names by number
275 ret = dbp[i]->open(dbp[i], NULL, dbfilename, NULL, DB_BTREE, DB_CREATE | DB_AUTO_COMMIT | DB_THREAD, 0600);
277 syslog(LOG_ERR, "bdb: db_open[%02x]: %s", i, db_strerror(ret));
279 syslog(LOG_ERR, "bdb: You may need to tune your database; please check http://www.citadel.org for more information.");
281 syslog(LOG_ERR, "bdb: exit code %d", ret);
288 // Close all of the db database files we've opened. This can be done in a loop, since it's just a bunch of closes.
289 void bdb_close_databases(void) {
293 static int closing = 0;
294 while (closing == 1) {
295 syslog(LOG_INFO, "bdb: already closing");
299 syslog(LOG_INFO, "bdb: performing final checkpoint");
300 if ((ret = dbenv->txn_checkpoint(dbenv, 0, 0, 0))) {
301 syslog(LOG_ERR, "bdb: txn_checkpoint: %s", db_strerror(ret));
304 syslog(LOG_INFO, "bdb: flushing the database logs");
305 if ((ret = dbenv->log_flush(dbenv, NULL))) {
306 syslog(LOG_ERR, "bdb: log_flush: %s", db_strerror(ret));
310 syslog(LOG_INFO, "bdb: closing databases");
311 for (i = 0; i < MAXCDB; ++i) {
312 syslog(LOG_INFO, "bdb: closing database %02x", i);
313 ret = dbp[i]->close(dbp[i], 0);
315 syslog(LOG_ERR, "bdb: db_close: %s", db_strerror(ret));
320 ret = dbenv->close(dbenv, DB_FORCESYNC);
322 syslog(LOG_ERR, "bdb: DBENV->close: %s", db_strerror(ret));
327 // Decompress a database item if it was compressed on disk
328 void bdb_decompress_if_necessary(struct cdbdata *cdb) {
329 static int magic = COMPRESS_MAGIC;
331 if ((cdb == NULL) || (cdb->ptr == NULL) || (cdb->len < sizeof(magic)) || (memcmp(cdb->ptr, &magic, sizeof(magic)))) {
335 // At this point we know we're looking at a compressed item.
337 struct CtdlCompressHeader zheader;
338 char *uncompressed_data;
339 char *compressed_data;
340 uLongf destLen, sourceLen;
343 memset(&zheader, 0, sizeof(struct CtdlCompressHeader));
344 cplen = sizeof(struct CtdlCompressHeader);
345 if (sizeof(struct CtdlCompressHeader) > cdb->len) {
348 memcpy(&zheader, cdb->ptr, cplen);
350 compressed_data = cdb->ptr;
351 compressed_data += sizeof(struct CtdlCompressHeader);
353 sourceLen = (uLongf) zheader.compressed_len;
354 destLen = (uLongf) zheader.uncompressed_len;
355 uncompressed_data = malloc(zheader.uncompressed_len);
357 if (uncompress((Bytef *) uncompressed_data,
358 (uLongf *) &destLen, (const Bytef *) compressed_data, (uLong) sourceLen) != Z_OK) {
359 syslog(LOG_ERR, "bdb: uncompress() error");
364 cdb->len = (size_t) destLen;
365 cdb->ptr = uncompressed_data;
369 // Store a piece of data. Returns 0 if the operation was successful. If a
370 // key already exists it should be overwritten.
371 int bdb_store(int cdb, const void *ckey, int ckeylen, void *cdata, int cdatalen) {
376 struct CtdlCompressHeader zheader;
377 char *compressed_data = NULL;
379 size_t buffer_len = 0;
382 memset(&dkey, 0, sizeof(DBT));
383 memset(&ddata, 0, sizeof(DBT));
385 dkey.data = (void *) ckey;
386 ddata.size = cdatalen;
389 // "visit" records are numerous and have big, mostly-empty string buffers in them.
390 // If we compress these we can get them down to 1% of their size most of the time.
391 if (cdb == CDB_VISIT) {
393 zheader.magic = COMPRESS_MAGIC;
394 zheader.uncompressed_len = cdatalen;
395 buffer_len = ((cdatalen * 101) / 100) + 100 + sizeof(struct CtdlCompressHeader);
396 destLen = (uLongf) buffer_len;
397 compressed_data = malloc(buffer_len);
398 if (compress2((Bytef *) (compressed_data + sizeof(struct CtdlCompressHeader)), &destLen, (Bytef *) cdata, (uLongf) cdatalen, 1) != Z_OK) {
399 syslog(LOG_ERR, "bdb: compress2() error");
402 zheader.compressed_len = (size_t) destLen;
403 memcpy(compressed_data, &zheader, sizeof(struct CtdlCompressHeader));
404 ddata.size = (size_t) (sizeof(struct CtdlCompressHeader) + zheader.compressed_len);
405 ddata.data = compressed_data;
408 if (TSD->tid != NULL) {
409 ret = dbp[cdb]->put(dbp[cdb], // db
410 TSD->tid, // transaction ID
416 syslog(LOG_ERR, "bdb: bdb_store(%d): %s", cdb, db_strerror(ret));
420 free(compressed_data);
425 bdb_bailIfCursor(TSD->cursors, "attempt to write during r/o cursor");
430 if ((ret = dbp[cdb]->put(dbp[cdb], // db
431 tid, // transaction ID
435 if (ret == DB_LOCK_DEADLOCK) {
440 syslog(LOG_ERR, "bdb: bdb_store(%d): %s", cdb, db_strerror(ret));
447 free(compressed_data);
456 // Delete a piece of data. Returns 0 if the operation was successful.
457 int bdb_delete(int cdb, void *key, int keylen) {
462 memset(&dkey, 0, sizeof dkey);
466 if (TSD->tid != NULL) {
467 ret = dbp[cdb]->del(dbp[cdb], TSD->tid, &dkey, 0);
469 syslog(LOG_ERR, "bdb: bdb_delete(%d): %s", cdb, db_strerror(ret));
470 if (ret != DB_NOTFOUND) {
476 bdb_bailIfCursor(TSD->cursors, "attempt to delete during r/o cursor");
481 if ((ret = dbp[cdb]->del(dbp[cdb], tid, &dkey, 0)) && ret != DB_NOTFOUND) {
482 if (ret == DB_LOCK_DEADLOCK) {
487 syslog(LOG_ERR, "bdb: bdb_delete(%d): %s", cdb, db_strerror(ret));
499 static DBC *bdb_localcursor(int cdb) {
503 if (TSD->cursors[cdb] == NULL) {
504 ret = dbp[cdb]->cursor(dbp[cdb], TSD->tid, &curs, 0);
507 ret = TSD->cursors[cdb]->c_dup(TSD->cursors[cdb], &curs, DB_POSITION);
511 syslog(LOG_ERR, "bdb: bdb_localcursor: %s", db_strerror(ret));
519 // Fetch a piece of data. If not found, returns NULL. Otherwise, it returns
520 // a struct cdbdata which it is the caller's responsibility to free later on
521 // using the bdb_free() routine.
522 struct cdbdata *bdb_fetch(int cdb, const void *key, int keylen) {
524 if (keylen == 0) { // key length zero is impossible
528 struct cdbdata *tempcdb;
532 memset(&dkey, 0, sizeof(DBT));
534 dkey.data = (void *) key;
536 if (TSD->tid != NULL) {
537 memset(&dret, 0, sizeof(DBT));
538 dret.flags = DB_DBT_MALLOC;
539 ret = dbp[cdb]->get(dbp[cdb], TSD->tid, &dkey, &dret, 0);
545 memset(&dret, 0, sizeof(DBT));
546 dret.flags = DB_DBT_MALLOC;
547 curs = bdb_localcursor(cdb);
548 ret = curs->c_get(curs, &dkey, &dret, DB_SET);
550 } while (ret == DB_LOCK_DEADLOCK);
553 if ((ret != 0) && (ret != DB_NOTFOUND)) {
554 syslog(LOG_ERR, "bdb: bdb_fetch(%d): %s", cdb, db_strerror(ret));
562 tempcdb = (struct cdbdata *) malloc(sizeof(struct cdbdata));
563 if (tempcdb == NULL) {
564 syslog(LOG_ERR, "bdb: bdb_fetch() cannot allocate memory for tempcdb: %m");
568 tempcdb->len = dret.size;
569 tempcdb->ptr = dret.data;
570 bdb_decompress_if_necessary(tempcdb);
576 // Free a cdbdata item.
578 // Note that we only free the 'ptr' portion if it is not NULL. This allows
579 // other code to assume ownership of that memory simply by storing the
580 // pointer elsewhere and then setting 'ptr' to NULL. bdb_free() will then
582 void bdb_free(struct cdbdata *cdb) {
590 void bdb_close_cursor(int cdb) {
591 if (TSD->cursors[cdb] != NULL) {
592 bdb_cclose(TSD->cursors[cdb]);
595 TSD->cursors[cdb] = NULL;
599 // Prepare for a sequential search of an entire database.
600 // (There is guaranteed to be no more than one traversal in progress per thread at any given time.)
601 void bdb_rewind(int cdb) {
604 if (TSD->cursors[cdb] != NULL) {
605 syslog(LOG_ERR, "bdb: bdb_rewind: must close cursor on database %d before reopening", cdb);
607 // bdb_cclose(TSD->cursors[cdb]);
610 // Now initialize the cursor
611 ret = dbp[cdb]->cursor(dbp[cdb], TSD->tid, &TSD->cursors[cdb], 0);
613 syslog(LOG_ERR, "bdb: bdb_rewind: db_cursor: %s", db_strerror(ret));
619 // Fetch the next item in a sequential search. Returns a pointer to a
620 // cdbdata structure, or NULL if we've hit the end.
621 struct cdbdata *bdb_next_item(int cdb) {
623 struct cdbdata *cdbret;
626 // Initialize the key/data pair so the flags aren't set.
627 memset(&key, 0, sizeof(key));
628 memset(&data, 0, sizeof(data));
629 data.flags = DB_DBT_MALLOC;
631 ret = TSD->cursors[cdb]->c_get(TSD->cursors[cdb], &key, &data, DB_NEXT);
634 if (ret != DB_NOTFOUND) {
635 syslog(LOG_ERR, "bdb: bdb_next_item(%d): %s", cdb, db_strerror(ret));
638 bdb_close_cursor(cdb);
639 return NULL; // presumably, end of file
642 cdbret = (struct cdbdata *) malloc(sizeof(struct cdbdata));
643 cdbret->len = data.size;
644 cdbret->ptr = data.data;
645 bdb_decompress_if_necessary(cdbret);
651 // Transaction-based stuff. I'm writing this as I bake cookies...
652 void bdb_begin_transaction(void) {
653 bdb_bailIfCursor(TSD->cursors, "can't begin transaction during r/o cursor");
655 if (TSD->tid != NULL) {
656 syslog(LOG_ERR, "bdb: bdb_begin_transaction: ERROR: nested transaction");
660 bdb_txbegin(&TSD->tid);
664 void bdb_end_transaction(void) {
667 for (i = 0; i < MAXCDB; i++) {
668 if (TSD->cursors[i] != NULL) {
669 syslog(LOG_WARNING, "bdb: bdb_end_transaction: WARNING: cursor %d still open at transaction end", i);
670 bdb_cclose(TSD->cursors[i]);
671 TSD->cursors[i] = NULL;
675 if (TSD->tid == NULL) {
676 syslog(LOG_ERR, "bdb: bdb_end_transaction: ERROR: bdb_txcommit(NULL) !!");
680 bdb_txcommit(TSD->tid);
687 // Truncate (delete every record)
688 void bdb_trunc(int cdb) {
692 if (TSD->tid != NULL) {
693 syslog(LOG_ERR, "bdb: bdb_trunc must not be called in a transaction.");
697 bdb_bailIfCursor(TSD->cursors, "attempt to write during r/o cursor");
701 if ((ret = dbp[cdb]->truncate(dbp[cdb], // db
702 NULL, // transaction ID
703 &count, // #rows deleted
705 if (ret == DB_LOCK_DEADLOCK) {
709 syslog(LOG_ERR, "bdb: bdb_truncate(%d): %s", cdb, db_strerror(ret));
711 syslog(LOG_ERR, "bdb: You may need to tune your database; please read http://www.citadel.org for more information.");
720 // compact (defragment) the database, possibly returning space back to the underlying filesystem
721 void bdb_compact(void) {
725 syslog(LOG_DEBUG, "bdb: bdb_compact() started");
726 for (i = 0; i < MAXCDB; i++) {
727 syslog(LOG_DEBUG, "bdb: compacting database %d", i);
728 ret = dbp[i]->compact(dbp[i], NULL, NULL, NULL, NULL, DB_FREE_SPACE, NULL);
730 syslog(LOG_ERR, "bdb: compact: %s", db_strerror(ret));
733 syslog(LOG_DEBUG, "bdb: bdb_compact() finished");
737 // Calling this function activates the Berkeley DB back end.
738 void bdb_init_backend(void) {
739 cdb_compact = bdb_compact;
740 cdb_checkpoint = bdb_checkpoint;
741 cdb_rewind = bdb_rewind;
742 cdb_fetch = bdb_fetch;
743 cdb_open_databases = bdb_open_databases;
744 cdb_close_databases = bdb_close_databases;
745 cdb_store = bdb_store;
746 cdb_delete = bdb_delete;
748 cdb_next_item = bdb_next_item;
749 cdb_close_cursor = bdb_close_cursor;
750 cdb_begin_transaction = bdb_begin_transaction;
751 cdb_end_transaction = bdb_end_transaction;
752 cdb_check_handles = bdb_check_handles;
753 cdb_trunc = bdb_trunc;
755 if (pthread_key_create(&bdb_thread_key, NULL) != 0) { // TSD key for this module
756 syslog(LOG_ERR, "pthread_key_create() : %m");
757 exit(CTDLEXIT_THREAD);
760 syslog(LOG_INFO, "db: initialized Berkeley DB backend");