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...
42 pthread_key_t bdb_thread_key;
43 #define TSD bdb_get_tsd()
46 // Some functions in this backend need to store some per-thread data.
47 // This returns the pointer to the current thread's per-thread data block, creating it if necessary.
48 struct bdb_tsd *bdb_get_tsd(void) {
50 struct bdb_tsd *c = (struct bdb_tsd *) pthread_getspecific(bdb_thread_key) ;
55 // If there's no TSD for this thread, it must be a new thread. Create our TSD region.
56 c = (struct bdb_tsd *) malloc(sizeof(struct bdb_tsd));
57 memset(c, 0, sizeof(struct bdb_tsd));
58 pthread_setspecific(bdb_thread_key, (const void *) c);
63 // Called by other functions in this module to GTFO quickly if we need to. Not part of the backend API.
64 void bdb_abort(void) {
65 syslog(LOG_DEBUG, "bdb: citserver is stopping in order to prevent data loss. uid=%d gid=%d euid=%d egid=%d",
66 getuid(), getgid(), geteuid(), getegid()
68 raise(SIGABRT); // This will exit in a way that can produce a core dump if needed.
69 exit(CTDLEXIT_DB); // Exit if the signal failed to end the program.
73 // Verbose logging callback for Berkeley DB. Not part of the backend API.
74 void bdb_verbose_log(const DB_ENV *bdb_env, const char *msg, const char *foo) {
75 if (!IsEmptyStr(msg)) {
76 syslog(LOG_DEBUG, "bdb: %s %s", msg, foo);
81 // Verbose error logging callback for Berkeley DB. Not part of the backend API.
82 void bdb_verbose_err(const DB_ENV *bdb_env, const char *errpfx, const char *msg) {
83 syslog(LOG_ERR, "bdb: %s", msg);
87 // Wrapper for txn_abort() that logs/aborts on error. Not part of the backend API.
88 static void bdb_txabort(DB_TXN *tid) {
91 ret = tid->abort(tid);
94 syslog(LOG_ERR, "bdb: txn_abort: %s", db_strerror(ret));
100 // Wrapper for txn_commit() that logs/aborts on error. Not part of the backend API.
101 static void bdb_txcommit(DB_TXN *tid) {
104 ret = tid->commit(tid, 0);
107 syslog(LOG_ERR, "bdb: txn_commit: %s", db_strerror(ret));
113 // Wrapper for txn_begin() that logs/aborts on error. Not part of the backend API.
114 static void bdb_txbegin(DB_TXN **tid) {
117 ret = bdb_env->txn_begin(bdb_env, NULL, tid, 0);
120 syslog(LOG_ERR, "bdb: txn_begin: %s", db_strerror(ret));
126 // Panic callback for Berkeley DB. Not part of the backend API.
127 static void bdb_dbpanic(DB_ENV *env, int errval) {
128 syslog(LOG_ERR, "bdb: PANIC: %s", db_strerror(errval));
133 // Close a cursor -- not part of the backend API.
134 static void bdb_cclose(DBC *cursor) {
137 if ((ret = cursor->c_close(cursor))) {
138 syslog(LOG_ERR, "bdb: c_close: %s", db_strerror(ret));
144 // Convenience function to abort if a cursor is still open when we didn't expect it to be.
145 // This is not part of the backend API.
146 static void bdb_bailIfCursor(DBC **cursors, const char *msg) {
149 for (i = 0; i < MAXCDB; i++)
150 if (cursors[i] != NULL) {
151 syslog(LOG_ERR, "bdb: cursor still in progress on cdb %02x: %s", i, msg);
157 void bdb_check_handles(void) {
158 bdb_bailIfCursor(TSD->cursors, "in check_handles");
160 if (TSD->tid != NULL) {
161 syslog(LOG_ERR, "bdb: transaction still in progress!");
167 // Request a checkpoint of the database. Called once per minute by the thread manager.
168 void bdb_checkpoint(void) {
171 syslog(LOG_DEBUG, "bdb: -- checkpoint --");
172 ret = bdb_env->txn_checkpoint(bdb_env, MAX_CHECKPOINT_KBYTES, MAX_CHECKPOINT_MINUTES, 0);
175 syslog(LOG_ERR, "bdb: bdb_checkpoint() txn_checkpoint: %s", db_strerror(ret));
179 // After a successful checkpoint, we can cull the unused logs
180 ret = bdb_env->log_set_config(bdb_env, DB_LOG_AUTO_REMOVE, 1);
182 syslog(LOG_ERR, "bdb: bdb_checkpoint() auto coll logs: %s", db_strerror(ret));
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 // 2023aug21 ajc: the third argument is zero, so this never did anything
229 // ret = bdb_env->set_cachesize(bdb_env, 0, 64 * 1024, 0);
231 // syslog(LOG_ERR, "bdb: set_cachesize: %s", db_strerror(ret));
232 // bdb_env->close(bdb_env, 0);
233 // syslog(LOG_ERR, "bdb: exit code %d", ret);
234 // exit(CTDLEXIT_DB);
237 if ((ret = bdb_env->set_lk_detect(bdb_env, DB_LOCK_DEFAULT))) {
238 syslog(LOG_ERR, "bdb: set_lk_detect: %s", db_strerror(ret));
239 bdb_env->close(bdb_env, 0);
240 syslog(LOG_ERR, "bdb: exit code %d", ret);
244 flags = DB_CREATE | DB_INIT_MPOOL | DB_PRIVATE | DB_INIT_TXN | DB_INIT_LOCK | DB_THREAD | DB_INIT_LOG;
245 syslog(LOG_DEBUG, "bdb: bdb_env->open(bdb_env, %s, %d, 0)", ctdl_db_dir, flags);
246 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try opening the database cleanly
247 if (ret == DB_RUNRECOVERY) {
248 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
249 syslog(LOG_ERR, "bdb: attempting recovery...");
251 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try recovery
253 if (ret == DB_RUNRECOVERY) {
254 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
255 syslog(LOG_ERR, "bdb: attempting catastrophic recovery...");
256 flags &= ~DB_RECOVER;
257 flags |= DB_RECOVER_FATAL;
258 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try catastrophic recovery
261 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
262 bdb_env->close(bdb_env, 0);
263 syslog(LOG_ERR, "bdb: exit code %d", ret);
267 syslog(LOG_INFO, "bdb: mounting databases");
268 for (i = 0; i < MAXCDB; ++i) {
269 ret = db_create(&bdb_table[i], bdb_env, 0); // Create a database handle
271 syslog(LOG_ERR, "bdb: db_create: %s", db_strerror(ret));
272 syslog(LOG_ERR, "bdb: exit code %d", ret);
276 snprintf(dbfilename, sizeof dbfilename, "cdb.%02x", i); // table names by number
277 ret = bdb_table[i]->open(bdb_table[i], NULL, dbfilename, NULL, DB_BTREE, DB_CREATE | DB_AUTO_COMMIT | DB_THREAD, 0600);
279 syslog(LOG_ERR, "bdb: db_open[%02x]: %s", i, db_strerror(ret));
281 syslog(LOG_ERR, "bdb: You may need to tune your database; please check http://www.citadel.org for more information.");
283 syslog(LOG_ERR, "bdb: exit code %d", ret);
290 // 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.
291 void bdb_close_databases(void) {
295 static int closing = 0;
296 while (closing == 1) {
297 syslog(LOG_INFO, "bdb: already closing");
301 syslog(LOG_INFO, "bdb: performing final checkpoint");
302 if ((ret = bdb_env->txn_checkpoint(bdb_env, 0, 0, 0))) {
303 syslog(LOG_ERR, "bdb: txn_checkpoint: %s", db_strerror(ret));
306 syslog(LOG_INFO, "bdb: flushing the database logs");
307 if ((ret = bdb_env->log_flush(bdb_env, NULL))) {
308 syslog(LOG_ERR, "bdb: log_flush: %s", db_strerror(ret));
312 syslog(LOG_INFO, "bdb: closing databases");
313 for (i = 0; i < MAXCDB; ++i) {
314 syslog(LOG_INFO, "bdb: closing database %02x", i);
315 ret = bdb_table[i]->close(bdb_table[i], 0);
317 syslog(LOG_ERR, "bdb: db_close: %s", db_strerror(ret));
322 ret = bdb_env->close(bdb_env, DB_FORCESYNC);
324 syslog(LOG_ERR, "bdb: DBENV->close: %s", db_strerror(ret));
329 // Decompress a DBT data block if it was compressed on disk.
330 void bdb_decompress_if_necessary(DBT *d) {
331 static int magic = COMPRESS_MAGIC;
333 if ((d == NULL) || (d->data == NULL) || (d->size < sizeof(magic)) || (memcmp(d->data, &magic, sizeof(magic)))) {
337 // At this point we know we're looking at a compressed item.
339 struct CtdlCompressHeader zheader;
340 char *uncompressed_data;
341 char *compressed_data;
342 uLongf destLen, sourceLen;
345 memset(&zheader, 0, sizeof(struct CtdlCompressHeader));
346 cplen = sizeof(struct CtdlCompressHeader);
347 if (sizeof(struct CtdlCompressHeader) > d->size) {
350 memcpy(&zheader, d->data, cplen);
352 compressed_data = d->data;
353 compressed_data += sizeof(struct CtdlCompressHeader);
355 sourceLen = (uLongf) zheader.compressed_len;
356 destLen = (uLongf) zheader.uncompressed_len;
357 uncompressed_data = malloc(zheader.uncompressed_len);
359 if (uncompress((Bytef *) uncompressed_data,
360 (uLongf *) &destLen, (const Bytef *) compressed_data, (uLong) sourceLen) != Z_OK) {
361 syslog(LOG_ERR, "bdb: uncompress() error");
366 d->size = (size_t) destLen;
367 d->data = uncompressed_data;
371 // Store a piece of data. Returns 0 if the operation was successful. If a
372 // key already exists it should be overwritten.
373 int bdb_store(int cdb, const void *ckey, int ckeylen, void *cdata, int cdatalen) {
378 struct CtdlCompressHeader zheader;
379 char *compressed_data = NULL;
381 size_t buffer_len = 0;
384 memset(&dkey, 0, sizeof(DBT));
385 memset(&ddata, 0, sizeof(DBT));
387 dkey.data = (void *) ckey;
388 ddata.size = cdatalen;
391 // "visit" records are numerous and have big, mostly-empty string buffers in them.
392 // If we compress these we can get them down to 1% of their size most of the time.
393 if (cdb == CDB_VISIT) {
395 zheader.magic = COMPRESS_MAGIC;
396 zheader.uncompressed_len = cdatalen;
397 buffer_len = ((cdatalen * 101) / 100) + 100 + sizeof(struct CtdlCompressHeader);
398 destLen = (uLongf) buffer_len;
399 compressed_data = malloc(buffer_len);
400 if (compress2((Bytef *) (compressed_data + sizeof(struct CtdlCompressHeader)), &destLen, (Bytef *) cdata, (uLongf) cdatalen, 1) != Z_OK) {
401 syslog(LOG_ERR, "bdb: compress2() error");
404 zheader.compressed_len = (size_t) destLen;
405 memcpy(compressed_data, &zheader, sizeof(struct CtdlCompressHeader));
406 ddata.size = (size_t) (sizeof(struct CtdlCompressHeader) + zheader.compressed_len);
407 ddata.data = compressed_data;
410 if (TSD->tid != NULL) {
411 ret = bdb_table[cdb]->put(bdb_table[cdb], // db
412 TSD->tid, // transaction ID
418 syslog(LOG_ERR, "bdb: bdb_store(%d): %s", cdb, db_strerror(ret));
422 free(compressed_data);
427 bdb_bailIfCursor(TSD->cursors, "attempt to write during r/o cursor");
432 if ((ret = bdb_table[cdb]->put(bdb_table[cdb], // db
433 tid, // transaction ID
437 if (ret == DB_LOCK_DEADLOCK) {
442 syslog(LOG_ERR, "bdb: bdb_store(%d): %s", cdb, db_strerror(ret));
449 free(compressed_data);
458 // Delete a piece of data. Returns 0 if the operation was successful.
459 int bdb_delete(int cdb, void *key, int keylen) {
464 memset(&dkey, 0, sizeof dkey);
468 if (TSD->tid != NULL) {
469 ret = bdb_table[cdb]->del(bdb_table[cdb], TSD->tid, &dkey, 0);
471 syslog(LOG_ERR, "bdb: bdb_delete(%d): %s", cdb, db_strerror(ret));
472 if (ret != DB_NOTFOUND) {
478 bdb_bailIfCursor(TSD->cursors, "attempt to delete during r/o cursor");
483 if ((ret = bdb_table[cdb]->del(bdb_table[cdb], tid, &dkey, 0)) && ret != DB_NOTFOUND) {
484 if (ret == DB_LOCK_DEADLOCK) {
489 syslog(LOG_ERR, "bdb: bdb_delete(%d): %s", cdb, db_strerror(ret));
501 static DBC *bdb_localcursor(int cdb) {
505 if (TSD->cursors[cdb] == NULL) {
506 ret = bdb_table[cdb]->cursor(bdb_table[cdb], TSD->tid, &curs, 0);
509 ret = TSD->cursors[cdb]->c_dup(TSD->cursors[cdb], &curs, DB_POSITION);
513 syslog(LOG_ERR, "bdb: bdb_localcursor: %s", db_strerror(ret));
521 // Fetch a piece of data. Returns a "struct cdbdata"
522 // cdbdata.len will be 0 if the item is not found.
523 struct cdbdata bdb_fetch(int cdb, const void *key, int keylen) {
525 struct cdbdata returned_data;
526 memset(&returned_data, 0, sizeof(struct cdbdata));
528 if (keylen == 0) { // key length zero is impossible
529 return(returned_data);
535 memset(&dkey, 0, sizeof(DBT));
537 dkey.data = (void *) key;
539 if (TSD->tid != NULL) {
540 TSD->dbdata[cdb].flags = DB_DBT_REALLOC;
541 ret = bdb_table[cdb]->get(bdb_table[cdb], TSD->tid, &dkey, &TSD->dbdata[cdb], 0);
547 TSD->dbdata[cdb].flags = DB_DBT_REALLOC;
548 curs = bdb_localcursor(cdb);
549 ret = curs->c_get(curs, &dkey, &TSD->dbdata[cdb], 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));
560 bdb_decompress_if_necessary(&TSD->dbdata[cdb]);
561 returned_data.len = TSD->dbdata[cdb].size;
562 returned_data.ptr = TSD->dbdata[cdb].data;
565 return(returned_data);
569 void bdb_close_cursor(int cdb) {
570 if (TSD->cursors[cdb] != NULL) {
571 bdb_cclose(TSD->cursors[cdb]);
574 TSD->cursors[cdb] = NULL;
578 // Prepare for a sequential search of an entire database.
579 // (There is guaranteed to be no more than one traversal in progress per thread at any given time.)
580 void bdb_rewind(int cdb) {
583 if (TSD->cursors[cdb] != NULL) {
584 syslog(LOG_ERR, "bdb: bdb_rewind: must close cursor on database %d before reopening", cdb);
586 // bdb_cclose(TSD->cursors[cdb]);
589 // Now initialize the cursor
590 ret = bdb_table[cdb]->cursor(bdb_table[cdb], TSD->tid, &TSD->cursors[cdb], 0);
592 syslog(LOG_ERR, "bdb: bdb_rewind: db_cursor: %s", db_strerror(ret));
598 // Fetch the next item in a sequential search. Returns a pointer to a
599 // cdbdata structure, or NULL if we've hit the end.
600 struct cdbkeyval bdb_next_item(int cdb) {
604 memset(&kv, 0, sizeof(struct cdbkeyval));
606 // reuse memory from the previous call.
607 TSD->dbkey[cdb].flags = DB_DBT_MALLOC;
608 TSD->dbdata[cdb].flags = DB_DBT_MALLOC;
610 ret = TSD->cursors[cdb]->c_get(TSD->cursors[cdb], &TSD->dbkey[cdb], &TSD->dbdata[cdb], DB_NEXT);
613 if (ret != DB_NOTFOUND) {
614 syslog(LOG_ERR, "bdb: bdb_next_item(%d): %s", cdb, db_strerror(ret));
617 bdb_close_cursor(cdb);
618 return(kv); // presumably, we are at the end
621 bdb_decompress_if_necessary(&TSD->dbdata[cdb]);
623 kv.key.len = TSD->dbkey[cdb].size;
624 kv.key.ptr = TSD->dbkey[cdb].data;
625 kv.val.len = TSD->dbdata[cdb].size;
626 kv.val.ptr = TSD->dbdata[cdb].data;
631 // Transaction-based stuff. I'm writing this as I bake cookies...
632 void bdb_begin_transaction(void) {
633 bdb_bailIfCursor(TSD->cursors, "can't begin transaction during r/o cursor");
635 if (TSD->tid != NULL) {
636 syslog(LOG_ERR, "bdb: bdb_begin_transaction: ERROR: nested transaction");
640 bdb_txbegin(&TSD->tid);
644 void bdb_end_transaction(void) {
647 for (i = 0; i < MAXCDB; i++) {
648 if (TSD->cursors[i] != NULL) {
649 syslog(LOG_WARNING, "bdb: bdb_end_transaction: WARNING: cursor %d still open at transaction end", i);
650 bdb_cclose(TSD->cursors[i]);
651 TSD->cursors[i] = NULL;
655 if (TSD->tid == NULL) {
656 syslog(LOG_ERR, "bdb: bdb_end_transaction: ERROR: bdb_txcommit(NULL) !!");
660 bdb_txcommit(TSD->tid);
667 // Truncate (delete every record)
668 void bdb_trunc(int cdb) {
672 if (TSD->tid != NULL) {
673 syslog(LOG_ERR, "bdb: bdb_trunc must not be called in a transaction.");
677 bdb_bailIfCursor(TSD->cursors, "attempt to write during r/o cursor");
681 if ((ret = bdb_table[cdb]->truncate(bdb_table[cdb], // db
682 NULL, // transaction ID
683 &count, // #rows deleted
685 if (ret == DB_LOCK_DEADLOCK) {
689 syslog(LOG_ERR, "bdb: bdb_truncate(%d): %s", cdb, db_strerror(ret));
691 syslog(LOG_ERR, "bdb: You may need to tune your database; please read http://www.citadel.org for more information.");
700 // compact (defragment) the database, possibly returning space back to the underlying filesystem
701 void bdb_compact(void) {
705 syslog(LOG_DEBUG, "bdb: bdb_compact() started");
706 for (i = 0; i < MAXCDB; i++) {
707 syslog(LOG_DEBUG, "bdb: compacting database %d", i);
708 ret = bdb_table[i]->compact(bdb_table[i], NULL, NULL, NULL, NULL, DB_FREE_SPACE, NULL);
710 syslog(LOG_ERR, "bdb: compact: %s", db_strerror(ret));
713 syslog(LOG_DEBUG, "bdb: bdb_compact() finished");
717 // Calling this function activates the Berkeley DB back end.
718 void bdb_init_backend(void) {
720 // Assign the backend API stubs to the functions in this module.
721 cdb_compact = bdb_compact;
722 cdb_checkpoint = bdb_checkpoint;
723 cdb_rewind = bdb_rewind;
724 cdb_fetch = bdb_fetch;
725 cdb_open_databases = bdb_open_databases;
726 cdb_close_databases = bdb_close_databases;
727 cdb_store = bdb_store;
728 cdb_delete = bdb_delete;
729 cdb_next_item = bdb_next_item;
730 cdb_close_cursor = bdb_close_cursor;
731 cdb_begin_transaction = bdb_begin_transaction;
732 cdb_end_transaction = bdb_end_transaction;
733 cdb_check_handles = bdb_check_handles;
734 cdb_trunc = bdb_trunc;
736 // Some functions in this backend need to store some per-thread data.
737 // We crerate the key here, during module initialization.
738 if (pthread_key_create(&bdb_thread_key, NULL) != 0) {
739 syslog(LOG_ERR, "pthread_key_create() : %m");
740 exit(CTDLEXIT_THREAD);
743 syslog(LOG_INFO, "db: initialized Berkeley DB backend");