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 *bdb_table[MAXCDB]; // One DB handle for each Citadel database
30 static DB_ENV *bdb_env; // The DB environment (global)
33 // These are items that we need to store "per thread" rather than "per session".
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 // Some functions in this backend need to store some per-thread data.
45 // This returns the pointer to the current thread's per-thread data block, creating it if necessary.
46 struct bdb_tsd *bdb_get_tsd(void) {
48 struct bdb_tsd *c = (struct bdb_tsd *) pthread_getspecific(bdb_thread_key) ;
53 // If there's no TSD for this thread, it must be a new thread. Create our TSD region.
54 c = (struct bdb_tsd *) malloc(sizeof(struct bdb_tsd));
55 memset(c, 0, sizeof(struct bdb_tsd));
56 pthread_setspecific(bdb_thread_key, (const void *) c);
61 // Called by other functions in this module to GTFO quickly if we need to. Not part of the backend API.
62 void bdb_abort(void) {
63 syslog(LOG_DEBUG, "bdb: citserver is stopping in order to prevent data loss. uid=%d gid=%d euid=%d egid=%d",
64 getuid(), getgid(), geteuid(), getegid()
66 raise(SIGABRT); // This will exit in a way that can produce a core dump if needed.
67 exit(CTDLEXIT_DB); // Exit if the signal failed to end the program.
71 // Verbose logging callback for Berkeley DB. Not part of the backend API.
72 void bdb_verbose_log(const DB_ENV *bdb_env, const char *msg, const char *foo) {
73 if (!IsEmptyStr(msg)) {
74 syslog(LOG_DEBUG, "bdb: %s %s", msg, foo);
79 // Verbose error logging callback for Berkeley DB. Not part of the backend API.
80 void bdb_verbose_err(const DB_ENV *bdb_env, const char *errpfx, const char *msg) {
81 syslog(LOG_ERR, "bdb: %s", msg);
85 // Wrapper for txn_abort() that logs/aborts on error. Not part of the backend API.
86 static void bdb_txabort(DB_TXN *tid) {
89 ret = tid->abort(tid);
92 syslog(LOG_ERR, "bdb: txn_abort: %s", db_strerror(ret));
98 // Wrapper for txn_commit() that logs/aborts on error. Not part of the backend API.
99 static void bdb_txcommit(DB_TXN *tid) {
102 ret = tid->commit(tid, 0);
105 syslog(LOG_ERR, "bdb: txn_commit: %s", db_strerror(ret));
111 // Wrapper for txn_begin() that logs/aborts on error. Not part of the backend API.
112 static void bdb_txbegin(DB_TXN **tid) {
115 ret = bdb_env->txn_begin(bdb_env, NULL, tid, 0);
118 syslog(LOG_ERR, "bdb: txn_begin: %s", db_strerror(ret));
124 // Panic callback for Berkeley DB. Not part of the backend API.
125 static void bdb_dbpanic(DB_ENV *env, int errval) {
126 syslog(LOG_ERR, "bdb: PANIC: %s", db_strerror(errval));
131 // Close a cursor -- not part of the backend API.
132 static void bdb_cclose(DBC *cursor) {
135 if ((ret = cursor->c_close(cursor))) {
136 syslog(LOG_ERR, "bdb: c_close: %s", db_strerror(ret));
142 // Convenience function to abort if a cursor is still open when we didn't expect it to be.
143 // This is not part of the backend API.
144 static void bdb_bailIfCursor(DBC **cursors, const char *msg) {
147 for (i = 0; i < MAXCDB; i++)
148 if (cursors[i] != NULL) {
149 syslog(LOG_ERR, "bdb: cursor still in progress on cdb %02x: %s", i, msg);
155 void bdb_check_handles(void) {
156 bdb_bailIfCursor(TSD->cursors, "in check_handles");
158 if (TSD->tid != NULL) {
159 syslog(LOG_ERR, "bdb: transaction still in progress!");
165 // Request a checkpoint of the database. Called once per minute by the thread manager.
166 void bdb_checkpoint(void) {
169 syslog(LOG_DEBUG, "bdb: -- checkpoint --");
170 ret = bdb_env->txn_checkpoint(bdb_env, MAX_CHECKPOINT_KBYTES, MAX_CHECKPOINT_MINUTES, 0);
173 syslog(LOG_ERR, "bdb: bdb_checkpoint() txn_checkpoint: %s", db_strerror(ret));
177 // After a successful checkpoint, we can cull the unused logs
178 if (CtdlGetConfigInt("c_auto_cull")) {
179 ret = bdb_env->log_set_config(bdb_env, DB_LOG_AUTO_REMOVE, 1);
182 ret = bdb_env->log_set_config(bdb_env, DB_LOG_AUTO_REMOVE, 0);
187 // Open the various tables we'll be using. Any table which
188 // does not exist should be created. Note that we don't need a
189 // critical section here, because there aren't any active threads
190 // manipulating the database yet.
191 void bdb_open_databases(void) {
196 int dbversion_major, dbversion_minor, dbversion_patch;
198 syslog(LOG_DEBUG, "bdb: bdb_open_databases() starting");
199 syslog(LOG_DEBUG, "bdb: Linked zlib: %s", zlibVersion());
200 syslog(LOG_DEBUG, "bdb: Compiled libdb: %s", DB_VERSION_STRING);
201 syslog(LOG_DEBUG, "bdb: Linked libdb: %s", db_version(&dbversion_major, &dbversion_minor, &dbversion_patch));
203 // Create synthetic integer version numbers and compare them.
204 // Never allow citserver to run with a libdb older then the one with which it was compiled.
205 int compiled_db_version = ( (DB_VERSION_MAJOR * 1000000) + (DB_VERSION_MINOR * 1000) + (DB_VERSION_PATCH) );
206 int linked_db_version = ( (dbversion_major * 1000000) + (dbversion_minor * 1000) + (dbversion_patch) );
207 if (compiled_db_version > linked_db_version) {
208 syslog(LOG_ERR, "bdb: citserver is running with a version of libdb older than the one with which it was compiled.");
209 syslog(LOG_ERR, "bdb: This is an invalid configuration. citserver will now exit to prevent data loss.");
213 syslog(LOG_DEBUG, "bdb: Setting up DB environment");
214 ret = db_env_create(&bdb_env, 0);
216 syslog(LOG_ERR, "bdb: db_env_create: %s", db_strerror(ret));
217 syslog(LOG_ERR, "bdb: exit code %d", ret);
220 bdb_env->set_errpfx(bdb_env, "citserver");
221 bdb_env->set_paniccall(bdb_env, bdb_dbpanic);
222 bdb_env->set_errcall(bdb_env, bdb_verbose_err);
223 bdb_env->set_msgcall(bdb_env, bdb_verbose_log);
224 bdb_env->set_verbose(bdb_env, DB_VERB_DEADLOCK, 1);
225 bdb_env->set_verbose(bdb_env, DB_VERB_RECOVERY, 1);
227 // We want to specify the shared memory buffer pool cachesize, but everything else is the default.
228 ret = bdb_env->set_cachesize(bdb_env, 0, 64 * 1024, 0);
230 syslog(LOG_ERR, "bdb: set_cachesize: %s", db_strerror(ret));
231 bdb_env->close(bdb_env, 0);
232 syslog(LOG_ERR, "bdb: exit code %d", ret);
236 if ((ret = bdb_env->set_lk_detect(bdb_env, DB_LOCK_DEFAULT))) {
237 syslog(LOG_ERR, "bdb: set_lk_detect: %s", db_strerror(ret));
238 bdb_env->close(bdb_env, 0);
239 syslog(LOG_ERR, "bdb: exit code %d", ret);
243 flags = DB_CREATE | DB_INIT_MPOOL | DB_PRIVATE | DB_INIT_TXN | DB_INIT_LOCK | DB_THREAD | DB_INIT_LOG;
244 syslog(LOG_DEBUG, "bdb: bdb_env->open(bdb_env, %s, %d, 0)", ctdl_db_dir, flags);
245 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try opening the database cleanly
246 if (ret == DB_RUNRECOVERY) {
247 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
248 syslog(LOG_ERR, "bdb: attempting recovery...");
250 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try recovery
252 if (ret == DB_RUNRECOVERY) {
253 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
254 syslog(LOG_ERR, "bdb: attempting catastrophic recovery...");
255 flags &= ~DB_RECOVER;
256 flags |= DB_RECOVER_FATAL;
257 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try catastrophic recovery
260 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
261 bdb_env->close(bdb_env, 0);
262 syslog(LOG_ERR, "bdb: exit code %d", ret);
266 syslog(LOG_INFO, "bdb: mounting databases");
267 for (i = 0; i < MAXCDB; ++i) {
268 ret = db_create(&bdb_table[i], bdb_env, 0); // Create a database handle
270 syslog(LOG_ERR, "bdb: db_create: %s", db_strerror(ret));
271 syslog(LOG_ERR, "bdb: exit code %d", ret);
275 snprintf(dbfilename, sizeof dbfilename, "cdb.%02x", i); // table names by number
276 ret = bdb_table[i]->open(bdb_table[i], NULL, dbfilename, NULL, DB_BTREE, DB_CREATE | DB_AUTO_COMMIT | DB_THREAD, 0600);
278 syslog(LOG_ERR, "bdb: db_open[%02x]: %s", i, db_strerror(ret));
280 syslog(LOG_ERR, "bdb: You may need to tune your database; please check http://www.citadel.org for more information.");
282 syslog(LOG_ERR, "bdb: exit code %d", ret);
289 // 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.
290 void bdb_close_databases(void) {
294 static int closing = 0;
295 while (closing == 1) {
296 syslog(LOG_INFO, "bdb: already closing");
300 syslog(LOG_INFO, "bdb: performing final checkpoint");
301 if ((ret = bdb_env->txn_checkpoint(bdb_env, 0, 0, 0))) {
302 syslog(LOG_ERR, "bdb: txn_checkpoint: %s", db_strerror(ret));
305 syslog(LOG_INFO, "bdb: flushing the database logs");
306 if ((ret = bdb_env->log_flush(bdb_env, NULL))) {
307 syslog(LOG_ERR, "bdb: log_flush: %s", db_strerror(ret));
311 syslog(LOG_INFO, "bdb: closing databases");
312 for (i = 0; i < MAXCDB; ++i) {
313 syslog(LOG_INFO, "bdb: closing database %02x", i);
314 ret = bdb_table[i]->close(bdb_table[i], 0);
316 syslog(LOG_ERR, "bdb: db_close: %s", db_strerror(ret));
321 ret = bdb_env->close(bdb_env, DB_FORCESYNC);
323 syslog(LOG_ERR, "bdb: DBENV->close: %s", db_strerror(ret));
328 // Decompress a database item if it was compressed on disk
329 void bdb_decompress_if_necessary(struct cdbdata *cdb) {
330 static int magic = COMPRESS_MAGIC;
332 if ((cdb == NULL) || (cdb->ptr == NULL) || (cdb->len < sizeof(magic)) || (memcmp(cdb->ptr, &magic, sizeof(magic)))) {
336 // At this point we know we're looking at a compressed item.
338 struct CtdlCompressHeader zheader;
339 char *uncompressed_data;
340 char *compressed_data;
341 uLongf destLen, sourceLen;
344 memset(&zheader, 0, sizeof(struct CtdlCompressHeader));
345 cplen = sizeof(struct CtdlCompressHeader);
346 if (sizeof(struct CtdlCompressHeader) > cdb->len) {
349 memcpy(&zheader, cdb->ptr, cplen);
351 compressed_data = cdb->ptr;
352 compressed_data += sizeof(struct CtdlCompressHeader);
354 sourceLen = (uLongf) zheader.compressed_len;
355 destLen = (uLongf) zheader.uncompressed_len;
356 uncompressed_data = malloc(zheader.uncompressed_len);
358 if (uncompress((Bytef *) uncompressed_data,
359 (uLongf *) &destLen, (const Bytef *) compressed_data, (uLong) sourceLen) != Z_OK) {
360 syslog(LOG_ERR, "bdb: uncompress() error");
365 cdb->len = (size_t) destLen;
366 cdb->ptr = uncompressed_data;
370 // Store a piece of data. Returns 0 if the operation was successful. If a
371 // key already exists it should be overwritten.
372 int bdb_store(int cdb, const void *ckey, int ckeylen, void *cdata, int cdatalen) {
377 struct CtdlCompressHeader zheader;
378 char *compressed_data = NULL;
380 size_t buffer_len = 0;
383 memset(&dkey, 0, sizeof(DBT));
384 memset(&ddata, 0, sizeof(DBT));
386 dkey.data = (void *) ckey;
387 ddata.size = cdatalen;
390 // "visit" records are numerous and have big, mostly-empty string buffers in them.
391 // If we compress these we can get them down to 1% of their size most of the time.
392 if (cdb == CDB_VISIT) {
394 zheader.magic = COMPRESS_MAGIC;
395 zheader.uncompressed_len = cdatalen;
396 buffer_len = ((cdatalen * 101) / 100) + 100 + sizeof(struct CtdlCompressHeader);
397 destLen = (uLongf) buffer_len;
398 compressed_data = malloc(buffer_len);
399 if (compress2((Bytef *) (compressed_data + sizeof(struct CtdlCompressHeader)), &destLen, (Bytef *) cdata, (uLongf) cdatalen, 1) != Z_OK) {
400 syslog(LOG_ERR, "bdb: compress2() error");
403 zheader.compressed_len = (size_t) destLen;
404 memcpy(compressed_data, &zheader, sizeof(struct CtdlCompressHeader));
405 ddata.size = (size_t) (sizeof(struct CtdlCompressHeader) + zheader.compressed_len);
406 ddata.data = compressed_data;
409 if (TSD->tid != NULL) {
410 ret = bdb_table[cdb]->put(bdb_table[cdb], // db
411 TSD->tid, // transaction ID
417 syslog(LOG_ERR, "bdb: bdb_store(%d): %s", cdb, db_strerror(ret));
421 free(compressed_data);
426 bdb_bailIfCursor(TSD->cursors, "attempt to write during r/o cursor");
431 if ((ret = bdb_table[cdb]->put(bdb_table[cdb], // db
432 tid, // transaction ID
436 if (ret == DB_LOCK_DEADLOCK) {
441 syslog(LOG_ERR, "bdb: bdb_store(%d): %s", cdb, db_strerror(ret));
448 free(compressed_data);
457 // Delete a piece of data. Returns 0 if the operation was successful.
458 int bdb_delete(int cdb, void *key, int keylen) {
463 memset(&dkey, 0, sizeof dkey);
467 if (TSD->tid != NULL) {
468 ret = bdb_table[cdb]->del(bdb_table[cdb], TSD->tid, &dkey, 0);
470 syslog(LOG_ERR, "bdb: bdb_delete(%d): %s", cdb, db_strerror(ret));
471 if (ret != DB_NOTFOUND) {
477 bdb_bailIfCursor(TSD->cursors, "attempt to delete during r/o cursor");
482 if ((ret = bdb_table[cdb]->del(bdb_table[cdb], tid, &dkey, 0)) && ret != DB_NOTFOUND) {
483 if (ret == DB_LOCK_DEADLOCK) {
488 syslog(LOG_ERR, "bdb: bdb_delete(%d): %s", cdb, db_strerror(ret));
500 static DBC *bdb_localcursor(int cdb) {
504 if (TSD->cursors[cdb] == NULL) {
505 ret = bdb_table[cdb]->cursor(bdb_table[cdb], TSD->tid, &curs, 0);
508 ret = TSD->cursors[cdb]->c_dup(TSD->cursors[cdb], &curs, DB_POSITION);
512 syslog(LOG_ERR, "bdb: bdb_localcursor: %s", db_strerror(ret));
520 // Fetch a piece of data. If not found, returns NULL. Otherwise, it returns
521 // a struct cdbdata which it is the caller's responsibility to free later on
522 // using the bdb_free() routine.
523 struct cdbdata *bdb_fetch(int cdb, const void *key, int keylen) {
525 if (keylen == 0) { // key length zero is impossible
529 struct cdbdata *tempcdb;
533 memset(&dkey, 0, sizeof(DBT));
535 dkey.data = (void *) key;
537 if (TSD->tid != NULL) {
538 memset(&dret, 0, sizeof(DBT));
539 dret.flags = DB_DBT_MALLOC;
540 ret = bdb_table[cdb]->get(bdb_table[cdb], TSD->tid, &dkey, &dret, 0);
546 memset(&dret, 0, sizeof(DBT));
547 dret.flags = DB_DBT_MALLOC;
548 curs = bdb_localcursor(cdb);
549 ret = curs->c_get(curs, &dkey, &dret, DB_SET);
551 } while (ret == DB_LOCK_DEADLOCK);
554 if ((ret != 0) && (ret != DB_NOTFOUND)) {
555 syslog(LOG_ERR, "bdb: bdb_fetch(%d): %s", cdb, db_strerror(ret));
563 tempcdb = (struct cdbdata *) malloc(sizeof(struct cdbdata));
564 if (tempcdb == NULL) {
565 syslog(LOG_ERR, "bdb: bdb_fetch() cannot allocate memory for tempcdb: %m");
569 tempcdb->len = dret.size;
570 tempcdb->ptr = dret.data;
571 bdb_decompress_if_necessary(tempcdb);
577 // Free a cdbdata item.
579 // Note that we only free the 'ptr' portion if it is not NULL. This allows
580 // other code to assume ownership of that memory simply by storing the
581 // pointer elsewhere and then setting 'ptr' to NULL. bdb_free() will then
583 void bdb_free(struct cdbdata *cdb) {
591 void bdb_close_cursor(int cdb) {
592 if (TSD->cursors[cdb] != NULL) {
593 bdb_cclose(TSD->cursors[cdb]);
596 TSD->cursors[cdb] = NULL;
600 // Prepare for a sequential search of an entire database.
601 // (There is guaranteed to be no more than one traversal in progress per thread at any given time.)
602 void bdb_rewind(int cdb) {
605 if (TSD->cursors[cdb] != NULL) {
606 syslog(LOG_ERR, "bdb: bdb_rewind: must close cursor on database %d before reopening", cdb);
608 // bdb_cclose(TSD->cursors[cdb]);
611 // Now initialize the cursor
612 ret = bdb_table[cdb]->cursor(bdb_table[cdb], TSD->tid, &TSD->cursors[cdb], 0);
614 syslog(LOG_ERR, "bdb: bdb_rewind: db_cursor: %s", db_strerror(ret));
620 // Fetch the next item in a sequential search. Returns a pointer to a
621 // cdbdata structure, or NULL if we've hit the end.
622 struct cdbdata *bdb_next_item(int cdb) {
624 struct cdbdata *cdbret;
627 // Initialize the key/data pair so the flags aren't set.
628 memset(&key, 0, sizeof(key));
629 memset(&data, 0, sizeof(data));
630 data.flags = DB_DBT_MALLOC;
632 ret = TSD->cursors[cdb]->c_get(TSD->cursors[cdb], &key, &data, DB_NEXT);
635 if (ret != DB_NOTFOUND) {
636 syslog(LOG_ERR, "bdb: bdb_next_item(%d): %s", cdb, db_strerror(ret));
639 bdb_close_cursor(cdb);
640 return NULL; // presumably, we are at the end
643 cdbret = (struct cdbdata *) malloc(sizeof(struct cdbdata));
644 cdbret->len = data.size;
645 cdbret->ptr = data.data;
646 bdb_decompress_if_necessary(cdbret);
652 // Transaction-based stuff. I'm writing this as I bake cookies...
653 void bdb_begin_transaction(void) {
654 bdb_bailIfCursor(TSD->cursors, "can't begin transaction during r/o cursor");
656 if (TSD->tid != NULL) {
657 syslog(LOG_ERR, "bdb: bdb_begin_transaction: ERROR: nested transaction");
661 bdb_txbegin(&TSD->tid);
665 void bdb_end_transaction(void) {
668 for (i = 0; i < MAXCDB; i++) {
669 if (TSD->cursors[i] != NULL) {
670 syslog(LOG_WARNING, "bdb: bdb_end_transaction: WARNING: cursor %d still open at transaction end", i);
671 bdb_cclose(TSD->cursors[i]);
672 TSD->cursors[i] = NULL;
676 if (TSD->tid == NULL) {
677 syslog(LOG_ERR, "bdb: bdb_end_transaction: ERROR: bdb_txcommit(NULL) !!");
681 bdb_txcommit(TSD->tid);
688 // Truncate (delete every record)
689 void bdb_trunc(int cdb) {
693 if (TSD->tid != NULL) {
694 syslog(LOG_ERR, "bdb: bdb_trunc must not be called in a transaction.");
698 bdb_bailIfCursor(TSD->cursors, "attempt to write during r/o cursor");
702 if ((ret = bdb_table[cdb]->truncate(bdb_table[cdb], // db
703 NULL, // transaction ID
704 &count, // #rows deleted
706 if (ret == DB_LOCK_DEADLOCK) {
710 syslog(LOG_ERR, "bdb: bdb_truncate(%d): %s", cdb, db_strerror(ret));
712 syslog(LOG_ERR, "bdb: You may need to tune your database; please read http://www.citadel.org for more information.");
721 // compact (defragment) the database, possibly returning space back to the underlying filesystem
722 void bdb_compact(void) {
726 syslog(LOG_DEBUG, "bdb: bdb_compact() started");
727 for (i = 0; i < MAXCDB; i++) {
728 syslog(LOG_DEBUG, "bdb: compacting database %d", i);
729 ret = bdb_table[i]->compact(bdb_table[i], NULL, NULL, NULL, NULL, DB_FREE_SPACE, NULL);
731 syslog(LOG_ERR, "bdb: compact: %s", db_strerror(ret));
734 syslog(LOG_DEBUG, "bdb: bdb_compact() finished");
738 // Calling this function activates the Berkeley DB back end.
739 void bdb_init_backend(void) {
741 // Assign the backend API stubs to the functions in this module.
742 cdb_compact = bdb_compact;
743 cdb_checkpoint = bdb_checkpoint;
744 cdb_rewind = bdb_rewind;
745 cdb_fetch = bdb_fetch;
746 cdb_open_databases = bdb_open_databases;
747 cdb_close_databases = bdb_close_databases;
748 cdb_store = bdb_store;
749 cdb_delete = bdb_delete;
751 cdb_next_item = bdb_next_item;
752 cdb_close_cursor = bdb_close_cursor;
753 cdb_begin_transaction = bdb_begin_transaction;
754 cdb_end_transaction = bdb_end_transaction;
755 cdb_check_handles = bdb_check_handles;
756 cdb_trunc = bdb_trunc;
758 // Some functions in this backend need to store some per-thread data.
759 // We crerate the key here, during module initialization.
760 if (pthread_key_create(&bdb_thread_key, NULL) != 0) {
761 syslog(LOG_ERR, "pthread_key_create() : %m");
762 exit(CTDLEXIT_THREAD);
765 syslog(LOG_INFO, "db: initialized Berkeley DB backend");