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"
20 #include <libcitadel.h>
21 #include "../../citserver.h"
22 #include "../../config.h"
23 #include "berkeley_db.h"
25 #if DB_VERSION_MAJOR < 18
26 #error Citadel requires Berkeley DB v18.0 or newer. Please upgrade.
29 // Globals (these are used across multiple functions in *this* module, but not elsewhere)
30 static DB *bdb_table[MAXCDB]; // One DB handle for each Citadel database
31 static DB_ENV *bdb_env; // The DB environment (global)
34 // These are items that we need to store "per thread" rather than "per session".
36 DB_TXN *tid; // Transaction handle
37 DBC *cursors[MAXCDB]; // Cursors, for traversals...
43 pthread_key_t bdb_thread_key;
44 #define TSD bdb_get_tsd()
47 // Some functions in this backend need to store some per-thread data.
48 // This returns the pointer to the current thread's per-thread data block, creating it if necessary.
49 // (This will also work in a non-threaded program; it will return the same pointer every time.)
50 struct bdb_tsd *bdb_get_tsd(void) {
52 struct bdb_tsd *c = (struct bdb_tsd *) pthread_getspecific(bdb_thread_key) ;
57 // If there's no TSD for this thread, it must be a new thread. Create our TSD region.
58 c = (struct bdb_tsd *) malloc(sizeof(struct bdb_tsd));
59 memset(c, 0, sizeof(struct bdb_tsd));
60 pthread_setspecific(bdb_thread_key, (const void *) c);
65 // Called by other functions in this module to GTFO quickly if we need to. Not part of the backend API.
66 void bdb_abort(void) {
67 syslog(LOG_DEBUG, "bdb: citserver is stopping in order to prevent data loss. uid=%d gid=%d euid=%d egid=%d",
68 getuid(), getgid(), geteuid(), getegid()
70 raise(SIGABRT); // This will exit in a way that can produce a core dump if needed.
71 exit(CTDLEXIT_DB); // Exit if the signal failed to end the program.
75 // Verbose logging callback for Berkeley DB. Not part of the backend API.
76 void bdb_verbose_log(const DB_ENV *bdb_env, const char *msg, const char *foo) {
77 if (!IsEmptyStr(msg)) {
78 syslog(LOG_DEBUG, "bdb: %s %s", msg, foo);
83 // Verbose error logging callback for Berkeley DB. Not part of the backend API.
84 void bdb_verbose_err(const DB_ENV *bdb_env, const char *errpfx, const char *msg) {
85 syslog(LOG_ERR, "bdb: %s", msg);
89 // Wrapper for txn_abort() that logs/aborts on error. Not part of the backend API.
90 static void bdb_txabort(DB_TXN *tid) {
93 ret = tid->abort(tid);
96 syslog(LOG_ERR, "bdb: txn_abort: %s", db_strerror(ret));
102 // Wrapper for txn_commit() that logs/aborts on error. Not part of the backend API.
103 static void bdb_txcommit(DB_TXN *tid) {
106 ret = tid->commit(tid, 0);
109 syslog(LOG_ERR, "bdb: txn_commit: %s", db_strerror(ret));
115 // Wrapper for txn_begin() that logs/aborts on error. Not part of the backend API.
116 static void bdb_txbegin(DB_TXN **tid) {
119 ret = bdb_env->txn_begin(bdb_env, NULL, tid, 0);
122 syslog(LOG_ERR, "bdb: txn_begin: %s", db_strerror(ret));
128 // Panic callback for Berkeley DB. Not part of the backend API.
129 static void bdb_dbpanic(DB_ENV *env, int errval) {
130 syslog(LOG_ERR, "bdb: PANIC: %s", db_strerror(errval));
135 // Close a cursor -- not part of the backend API.
136 static void bdb_cclose(DBC *cursor) {
139 if ((ret = cursor->c_close(cursor))) {
140 syslog(LOG_ERR, "bdb: c_close: %s", db_strerror(ret));
146 // Convenience function to abort if a cursor is still open when we didn't expect it to be.
147 // This is not part of the backend API.
148 static void bdb_bailIfCursor(DBC **cursors, const char *msg) {
151 for (i = 0; i < MAXCDB; i++)
152 if (cursors[i] != NULL) {
153 syslog(LOG_ERR, "bdb: cursor still in progress on cdb %02x: %s", i, msg);
159 void bdb_check_handles(void) {
160 bdb_bailIfCursor(TSD->cursors, "in check_handles");
162 if (TSD->tid != NULL) {
163 syslog(LOG_ERR, "bdb: transaction still in progress!");
169 // Request a checkpoint of the database. Called once per minute by the thread manager.
170 void bdb_checkpoint(void) {
173 syslog(LOG_DEBUG, "bdb: -- checkpoint --");
174 ret = bdb_env->txn_checkpoint(bdb_env, MAX_CHECKPOINT_KBYTES, MAX_CHECKPOINT_MINUTES, 0);
177 syslog(LOG_ERR, "bdb: bdb_checkpoint() txn_checkpoint: %s", db_strerror(ret));
181 // After a successful checkpoint, we can cull the unused logs
182 ret = bdb_env->log_set_config(bdb_env, DB_LOG_AUTO_REMOVE, 1);
184 syslog(LOG_ERR, "bdb: bdb_checkpoint() auto coll logs: %s", db_strerror(ret));
189 // Open the various tables we'll be using. Any table which
190 // does not exist should be created. Note that we don't need a
191 // critical section here, because there aren't any active threads
192 // manipulating the database yet.
193 void bdb_open_databases(void) {
198 int dbversion_major, dbversion_minor, dbversion_patch;
200 syslog(LOG_DEBUG, "bdb: bdb_open_databases() starting");
201 syslog(LOG_DEBUG, "bdb: Linked zlib: %s", zlibVersion());
202 syslog(LOG_DEBUG, "bdb: Compiled libdb: %s", DB_VERSION_STRING);
203 syslog(LOG_DEBUG, "bdb: Linked libdb: %s", db_version(&dbversion_major, &dbversion_minor, &dbversion_patch));
205 // Create synthetic integer version numbers and compare them.
206 // Never allow citserver to run with a libdb older then the one with which it was compiled.
207 int compiled_db_version = ( (DB_VERSION_MAJOR * 1000000) + (DB_VERSION_MINOR * 1000) + (DB_VERSION_PATCH) );
208 int linked_db_version = ( (dbversion_major * 1000000) + (dbversion_minor * 1000) + (dbversion_patch) );
209 if (compiled_db_version > linked_db_version) {
210 syslog(LOG_ERR, "bdb: citserver is running with a version of libdb older than the one with which it was compiled.");
211 syslog(LOG_ERR, "bdb: This is an invalid configuration. citserver will now exit to prevent data loss.");
215 syslog(LOG_DEBUG, "bdb: Setting up DB environment");
216 ret = db_env_create(&bdb_env, 0);
218 syslog(LOG_ERR, "bdb: db_env_create: %s", db_strerror(ret));
219 syslog(LOG_ERR, "bdb: exit code %d", ret);
222 bdb_env->set_errpfx(bdb_env, "citserver");
223 bdb_env->set_paniccall(bdb_env, bdb_dbpanic);
224 bdb_env->set_errcall(bdb_env, bdb_verbose_err);
225 bdb_env->set_msgcall(bdb_env, bdb_verbose_log);
226 bdb_env->set_verbose(bdb_env, DB_VERB_DEADLOCK, 1);
227 bdb_env->set_verbose(bdb_env, DB_VERB_RECOVERY, 1);
229 // We want to specify the shared memory buffer pool cachesize, but everything else is the default.
230 // 2023aug21 ajc: the third argument is zero, so this never did anything
231 // ret = bdb_env->set_cachesize(bdb_env, 0, 64 * 1024, 0);
233 // syslog(LOG_ERR, "bdb: set_cachesize: %s", db_strerror(ret));
234 // bdb_env->close(bdb_env, 0);
235 // syslog(LOG_ERR, "bdb: exit code %d", ret);
236 // exit(CTDLEXIT_DB);
239 if ((ret = bdb_env->set_lk_detect(bdb_env, DB_LOCK_DEFAULT))) {
240 syslog(LOG_ERR, "bdb: set_lk_detect: %s", db_strerror(ret));
241 bdb_env->close(bdb_env, 0);
242 syslog(LOG_ERR, "bdb: exit code %d", ret);
246 flags = DB_CREATE | DB_INIT_MPOOL | DB_PRIVATE | DB_INIT_TXN | DB_INIT_LOCK | DB_THREAD | DB_INIT_LOG;
247 syslog(LOG_DEBUG, "bdb: bdb_env->open(bdb_env, %s, %d, 0)", ctdl_db_dir, flags);
248 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try opening the database cleanly
249 if (ret == DB_RUNRECOVERY) {
250 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
251 syslog(LOG_ERR, "bdb: attempting recovery...");
253 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try recovery
255 if (ret == DB_RUNRECOVERY) {
256 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
257 syslog(LOG_ERR, "bdb: attempting catastrophic recovery...");
258 flags &= ~DB_RECOVER;
259 flags |= DB_RECOVER_FATAL;
260 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try catastrophic recovery
263 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
264 bdb_env->close(bdb_env, 0);
265 syslog(LOG_ERR, "bdb: exit code %d", ret);
269 for (i = 0; i < MAXCDB; ++i) {
270 syslog(LOG_INFO, "bdb: mounting database %02x", i);
271 ret = db_create(&bdb_table[i], bdb_env, 0); // Create a database handle
273 syslog(LOG_ERR, "bdb: db_create: %s", db_strerror(ret));
274 syslog(LOG_ERR, "bdb: exit code %d", ret);
278 snprintf(dbfilename, sizeof dbfilename, "cdb.%02x", i); // table names by number
279 ret = bdb_table[i]->open(bdb_table[i], NULL, dbfilename, NULL, DB_BTREE, DB_CREATE | DB_AUTO_COMMIT | DB_THREAD, 0600);
281 syslog(LOG_ERR, "bdb: db_open[%02x]: %s", i, db_strerror(ret));
283 syslog(LOG_ERR, "bdb: You may need to tune your database; please check http://www.citadel.org for more information.");
285 syslog(LOG_ERR, "bdb: exit code %d", ret);
292 // 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.
293 void bdb_close_databases(void) {
297 static int closing = 0;
298 while (closing == 1) {
299 syslog(LOG_INFO, "bdb: already closing");
303 syslog(LOG_INFO, "bdb: performing final checkpoint");
304 if ((ret = bdb_env->txn_checkpoint(bdb_env, 0, 0, 0))) {
305 syslog(LOG_ERR, "bdb: txn_checkpoint: %s", db_strerror(ret));
308 syslog(LOG_INFO, "bdb: flushing the database logs");
309 if ((ret = bdb_env->log_flush(bdb_env, NULL))) {
310 syslog(LOG_ERR, "bdb: log_flush: %s", db_strerror(ret));
314 syslog(LOG_INFO, "bdb: closing databases");
315 for (i = 0; i < MAXCDB; ++i) {
316 syslog(LOG_INFO, "bdb: closing database %02x", i);
317 ret = bdb_table[i]->close(bdb_table[i], 0);
319 syslog(LOG_ERR, "bdb: db_close: %s", db_strerror(ret));
324 ret = bdb_env->close(bdb_env, DB_FORCESYNC);
326 syslog(LOG_ERR, "bdb: DBENV->close: %s", db_strerror(ret));
331 // Decompress a DBT data block if it was compressed on disk.
332 void bdb_decompress_if_necessary(DBT *d) {
333 static int magic = COMPRESS_MAGIC;
335 if ((d == NULL) || (d->data == NULL) || (d->size < sizeof(magic)) || (memcmp(d->data, &magic, sizeof(magic)))) {
339 // At this point we know we're looking at a compressed item.
341 struct CtdlCompressHeader zheader;
342 char *uncompressed_data;
343 char *compressed_data;
344 uLongf destLen, sourceLen;
347 memset(&zheader, 0, sizeof(struct CtdlCompressHeader));
348 cplen = sizeof(struct CtdlCompressHeader);
349 if (sizeof(struct CtdlCompressHeader) > d->size) {
352 memcpy(&zheader, d->data, cplen);
354 compressed_data = d->data;
355 compressed_data += sizeof(struct CtdlCompressHeader);
357 sourceLen = (uLongf) zheader.compressed_len;
358 destLen = (uLongf) zheader.uncompressed_len;
359 uncompressed_data = malloc(zheader.uncompressed_len);
361 if (uncompress((Bytef *) uncompressed_data,
362 (uLongf *) &destLen, (const Bytef *) compressed_data, (uLong) sourceLen) != Z_OK) {
363 syslog(LOG_ERR, "bdb: uncompress() error");
368 d->size = (size_t) destLen;
369 d->data = uncompressed_data;
373 // Store a piece of data. Returns 0 if the operation was successful. If a
374 // key already exists it should be overwritten.
375 int bdb_store(int cdb, const void *ckey, int ckeylen, void *cdata, int cdatalen) {
380 struct CtdlCompressHeader zheader;
381 char *compressed_data = NULL;
383 size_t buffer_len = 0;
386 memset(&dkey, 0, sizeof(DBT));
387 memset(&ddata, 0, sizeof(DBT));
389 dkey.data = (void *) ckey;
390 ddata.size = cdatalen;
393 // "visit" records are numerous and have big, mostly-empty string buffers in them.
394 // If we compress these we can get them down to 1% of their size most of the time.
395 if (cdb == CDB_VISIT) {
397 zheader.magic = COMPRESS_MAGIC;
398 zheader.uncompressed_len = cdatalen;
399 buffer_len = ((cdatalen * 101) / 100) + 100 + sizeof(struct CtdlCompressHeader);
400 destLen = (uLongf) buffer_len;
401 compressed_data = malloc(buffer_len);
402 if (compress2((Bytef *) (compressed_data + sizeof(struct CtdlCompressHeader)), &destLen, (Bytef *) cdata, (uLongf) cdatalen, 1) != Z_OK) {
403 syslog(LOG_ERR, "bdb: compress2() error");
406 zheader.compressed_len = (size_t) destLen;
407 memcpy(compressed_data, &zheader, sizeof(struct CtdlCompressHeader));
408 ddata.size = (size_t) (sizeof(struct CtdlCompressHeader) + zheader.compressed_len);
409 ddata.data = compressed_data;
412 if (TSD->tid != NULL) {
413 ret = bdb_table[cdb]->put(bdb_table[cdb], // db
414 TSD->tid, // transaction ID
420 syslog(LOG_ERR, "bdb: bdb_store(%d): %s", cdb, db_strerror(ret));
424 free(compressed_data);
429 bdb_bailIfCursor(TSD->cursors, "attempt to write during r/o cursor");
434 if ((ret = bdb_table[cdb]->put(bdb_table[cdb], // db
435 tid, // transaction ID
439 if (ret == DB_LOCK_DEADLOCK) {
444 syslog(LOG_ERR, "bdb: bdb_store(%d): %s", cdb, db_strerror(ret));
451 free(compressed_data);
460 // Delete a piece of data. Returns 0 if the operation was successful.
461 int bdb_delete(int cdb, void *key, int keylen) {
466 memset(&dkey, 0, sizeof dkey);
470 if (TSD->tid != NULL) {
471 ret = bdb_table[cdb]->del(bdb_table[cdb], TSD->tid, &dkey, 0);
473 syslog(LOG_ERR, "bdb: bdb_delete(%d): %s", cdb, db_strerror(ret));
474 if (ret != DB_NOTFOUND) {
480 bdb_bailIfCursor(TSD->cursors, "attempt to delete during r/o cursor");
485 if ((ret = bdb_table[cdb]->del(bdb_table[cdb], tid, &dkey, 0)) && ret != DB_NOTFOUND) {
486 if (ret == DB_LOCK_DEADLOCK) {
491 syslog(LOG_ERR, "bdb: bdb_delete(%d): %s", cdb, db_strerror(ret));
503 static DBC *bdb_localcursor(int cdb) {
507 if (TSD->cursors[cdb] == NULL) {
508 ret = bdb_table[cdb]->cursor(bdb_table[cdb], TSD->tid, &curs, 0);
511 ret = TSD->cursors[cdb]->c_dup(TSD->cursors[cdb], &curs, DB_POSITION);
515 syslog(LOG_ERR, "bdb: bdb_localcursor: %s", db_strerror(ret));
523 // Fetch a piece of data. Returns a "struct cdbdata"
524 // cdbdata.len will be 0 if the item is not found.
525 struct cdbdata bdb_fetch(int cdb, const void *key, int keylen) {
527 struct cdbdata returned_data;
528 memset(&returned_data, 0, sizeof(struct cdbdata));
530 if (keylen == 0) { // key length zero is impossible
531 return(returned_data);
537 memset(&dkey, 0, sizeof(DBT));
539 dkey.data = (void *) key;
541 if (TSD->tid != NULL) {
542 TSD->dbdata[cdb].flags = DB_DBT_REALLOC;
543 ret = bdb_table[cdb]->get(bdb_table[cdb], TSD->tid, &dkey, &TSD->dbdata[cdb], 0);
549 TSD->dbdata[cdb].flags = DB_DBT_REALLOC;
550 curs = bdb_localcursor(cdb);
551 ret = curs->c_get(curs, &dkey, &TSD->dbdata[cdb], DB_SET);
553 } while (ret == DB_LOCK_DEADLOCK);
556 if ((ret != 0) && (ret != DB_NOTFOUND)) {
557 syslog(LOG_ERR, "bdb: bdb_fetch(%d): %s", cdb, db_strerror(ret));
562 bdb_decompress_if_necessary(&TSD->dbdata[cdb]);
563 returned_data.len = TSD->dbdata[cdb].size;
564 returned_data.ptr = TSD->dbdata[cdb].data;
567 return(returned_data);
571 void bdb_close_cursor(int cdb) {
572 if (TSD->cursors[cdb] != NULL) {
573 bdb_cclose(TSD->cursors[cdb]);
576 TSD->cursors[cdb] = NULL;
580 // Prepare for a sequential search of an entire database.
581 // (There is guaranteed to be no more than one traversal in progress per thread at any given time.)
582 void bdb_rewind(int cdb) {
585 if (TSD->cursors[cdb] != NULL) {
586 syslog(LOG_ERR, "bdb: bdb_rewind: must close cursor on database %d before reopening", cdb);
588 // bdb_cclose(TSD->cursors[cdb]);
591 // Now initialize the cursor
592 ret = bdb_table[cdb]->cursor(bdb_table[cdb], TSD->tid, &TSD->cursors[cdb], 0);
594 syslog(LOG_ERR, "bdb: bdb_rewind: db_cursor: %s", db_strerror(ret));
600 // Fetch the next item in a sequential search. Returns a pointer to a
601 // cdbdata structure, or NULL if we've hit the end.
602 struct cdbkeyval bdb_next_item(int cdb) {
606 memset(&kv, 0, sizeof(struct cdbkeyval));
608 // reuse memory from the previous call.
609 TSD->dbkey[cdb].flags = DB_DBT_MALLOC;
610 TSD->dbdata[cdb].flags = DB_DBT_MALLOC;
612 assert(TSD->cursors[cdb] != NULL);
613 ret = TSD->cursors[cdb]->c_get(TSD->cursors[cdb], &TSD->dbkey[cdb], &TSD->dbdata[cdb], DB_NEXT);
616 if (ret != DB_NOTFOUND) {
617 syslog(LOG_ERR, "bdb: bdb_next_item(%d): %s", cdb, db_strerror(ret));
620 bdb_close_cursor(cdb);
621 return(kv); // presumably, we are at the end
624 bdb_decompress_if_necessary(&TSD->dbdata[cdb]);
626 kv.key.len = TSD->dbkey[cdb].size;
627 kv.key.ptr = TSD->dbkey[cdb].data;
628 kv.val.len = TSD->dbdata[cdb].size;
629 kv.val.ptr = TSD->dbdata[cdb].data;
634 // Transaction-based stuff. I'm writing this as I bake cookies...
635 void bdb_begin_transaction(void) {
636 bdb_bailIfCursor(TSD->cursors, "can't begin transaction during r/o cursor");
638 if (TSD->tid != NULL) {
639 syslog(LOG_ERR, "bdb: bdb_begin_transaction: ERROR: nested transaction");
643 bdb_txbegin(&TSD->tid);
647 void bdb_end_transaction(void) {
650 for (i = 0; i < MAXCDB; i++) {
651 if (TSD->cursors[i] != NULL) {
652 syslog(LOG_WARNING, "bdb: bdb_end_transaction: WARNING: cursor %d still open at transaction end", i);
653 bdb_cclose(TSD->cursors[i]);
654 TSD->cursors[i] = NULL;
658 if (TSD->tid == NULL) {
659 syslog(LOG_ERR, "bdb: bdb_end_transaction: ERROR: bdb_txcommit(NULL) !!");
663 bdb_txcommit(TSD->tid);
670 // Truncate (delete every record)
671 void bdb_trunc(int cdb) {
675 if (TSD->tid != NULL) {
676 syslog(LOG_ERR, "bdb: bdb_trunc must not be called in a transaction.");
680 bdb_bailIfCursor(TSD->cursors, "attempt to write during r/o cursor");
684 if ((ret = bdb_table[cdb]->truncate(bdb_table[cdb], // db
685 NULL, // transaction ID
686 &count, // #rows deleted
688 if (ret == DB_LOCK_DEADLOCK) {
692 syslog(LOG_ERR, "bdb: bdb_truncate(%d): %s", cdb, db_strerror(ret));
694 syslog(LOG_ERR, "bdb: You may need to tune your database; please read http://www.citadel.org for more information.");
703 // compact (defragment) the database, possibly returning space back to the underlying filesystem
704 void bdb_compact(void) {
708 syslog(LOG_DEBUG, "bdb: bdb_compact() started");
709 for (i = 0; i < MAXCDB; i++) {
710 syslog(LOG_DEBUG, "bdb: compacting database %d", i);
711 ret = bdb_table[i]->compact(bdb_table[i], NULL, NULL, NULL, NULL, DB_FREE_SPACE, NULL);
713 syslog(LOG_ERR, "bdb: compact: %s", db_strerror(ret));
716 syslog(LOG_DEBUG, "bdb: bdb_compact() finished");
720 // Calling this function activates the Berkeley DB back end.
721 void bdb_init_backend(void) {
723 // Assign the backend API stubs to the functions in this module.
724 cdb_compact = bdb_compact;
725 cdb_checkpoint = bdb_checkpoint;
726 cdb_rewind = bdb_rewind;
727 cdb_fetch = bdb_fetch;
728 cdb_open_databases = bdb_open_databases;
729 cdb_close_databases = bdb_close_databases;
730 cdb_store = bdb_store;
731 cdb_delete = bdb_delete;
732 cdb_next_item = bdb_next_item;
733 cdb_close_cursor = bdb_close_cursor;
734 cdb_begin_transaction = bdb_begin_transaction;
735 cdb_end_transaction = bdb_end_transaction;
736 cdb_check_handles = bdb_check_handles;
737 cdb_trunc = bdb_trunc;
739 // Some functions in this backend need to store some per-thread data.
740 // We crerate the key here, during module initialization.
741 if (pthread_key_create(&bdb_thread_key, NULL) != 0) {
742 syslog(LOG_ERR, "pthread_key_create() : %m");
743 exit(CTDLEXIT_THREAD);
746 syslog(LOG_INFO, "db: initialized Berkeley DB backend");