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 if (CtdlGetConfigInt("c_auto_cull")) {
181 ret = bdb_env->log_set_config(bdb_env, DB_LOG_AUTO_REMOVE, 1);
184 ret = bdb_env->log_set_config(bdb_env, DB_LOG_AUTO_REMOVE, 0);
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 ret = bdb_env->set_cachesize(bdb_env, 0, 64 * 1024, 0);
232 syslog(LOG_ERR, "bdb: set_cachesize: %s", db_strerror(ret));
233 bdb_env->close(bdb_env, 0);
234 syslog(LOG_ERR, "bdb: exit code %d", ret);
238 if ((ret = bdb_env->set_lk_detect(bdb_env, DB_LOCK_DEFAULT))) {
239 syslog(LOG_ERR, "bdb: set_lk_detect: %s", db_strerror(ret));
240 bdb_env->close(bdb_env, 0);
241 syslog(LOG_ERR, "bdb: exit code %d", ret);
245 flags = DB_CREATE | DB_INIT_MPOOL | DB_PRIVATE | DB_INIT_TXN | DB_INIT_LOCK | DB_THREAD | DB_INIT_LOG;
246 syslog(LOG_DEBUG, "bdb: bdb_env->open(bdb_env, %s, %d, 0)", ctdl_db_dir, flags);
247 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try opening the database cleanly
248 if (ret == DB_RUNRECOVERY) {
249 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
250 syslog(LOG_ERR, "bdb: attempting recovery...");
252 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try recovery
254 if (ret == DB_RUNRECOVERY) {
255 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
256 syslog(LOG_ERR, "bdb: attempting catastrophic recovery...");
257 flags &= ~DB_RECOVER;
258 flags |= DB_RECOVER_FATAL;
259 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try catastrophic recovery
262 syslog(LOG_ERR, "bdb: bdb_env->open: %s", db_strerror(ret));
263 bdb_env->close(bdb_env, 0);
264 syslog(LOG_ERR, "bdb: exit code %d", ret);
268 syslog(LOG_INFO, "bdb: mounting databases");
269 for (i = 0; i < MAXCDB; ++i) {
270 ret = db_create(&bdb_table[i], bdb_env, 0); // Create a database handle
272 syslog(LOG_ERR, "bdb: db_create: %s", db_strerror(ret));
273 syslog(LOG_ERR, "bdb: exit code %d", ret);
277 snprintf(dbfilename, sizeof dbfilename, "cdb.%02x", i); // table names by number
278 ret = bdb_table[i]->open(bdb_table[i], NULL, dbfilename, NULL, DB_BTREE, DB_CREATE | DB_AUTO_COMMIT | DB_THREAD, 0600);
280 syslog(LOG_ERR, "bdb: db_open[%02x]: %s", i, db_strerror(ret));
282 syslog(LOG_ERR, "bdb: You may need to tune your database; please check http://www.citadel.org for more information.");
284 syslog(LOG_ERR, "bdb: exit code %d", ret);
291 // 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.
292 void bdb_close_databases(void) {
296 static int closing = 0;
297 while (closing == 1) {
298 syslog(LOG_INFO, "bdb: already closing");
302 syslog(LOG_INFO, "bdb: performing final checkpoint");
303 if ((ret = bdb_env->txn_checkpoint(bdb_env, 0, 0, 0))) {
304 syslog(LOG_ERR, "bdb: txn_checkpoint: %s", db_strerror(ret));
307 syslog(LOG_INFO, "bdb: flushing the database logs");
308 if ((ret = bdb_env->log_flush(bdb_env, NULL))) {
309 syslog(LOG_ERR, "bdb: log_flush: %s", db_strerror(ret));
313 syslog(LOG_INFO, "bdb: closing databases");
314 for (i = 0; i < MAXCDB; ++i) {
315 syslog(LOG_INFO, "bdb: closing database %02x", i);
316 ret = bdb_table[i]->close(bdb_table[i], 0);
318 syslog(LOG_ERR, "bdb: db_close: %s", db_strerror(ret));
323 ret = bdb_env->close(bdb_env, DB_FORCESYNC);
325 syslog(LOG_ERR, "bdb: DBENV->close: %s", db_strerror(ret));
330 // Decompress a database item if it was compressed on disk
331 void bdb_decompress_if_necessary(struct cdbdata *cdb) {
332 static int magic = COMPRESS_MAGIC;
334 if ((cdb == NULL) || (cdb->ptr == NULL) || (cdb->len < sizeof(magic)) || (memcmp(cdb->ptr, &magic, sizeof(magic)))) {
335 cdb->cdbfree_must_free_ptr = 0;
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) > cdb->len) {
352 memcpy(&zheader, cdb->ptr, cplen);
354 compressed_data = cdb->ptr;
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 cdb->len = (size_t) destLen;
369 cdb->ptr = uncompressed_data;
370 cdb->cdbfree_must_free_ptr = 1;
374 // Store a piece of data. Returns 0 if the operation was successful. If a
375 // key already exists it should be overwritten.
376 int bdb_store(int cdb, const void *ckey, int ckeylen, void *cdata, int cdatalen) {
381 struct CtdlCompressHeader zheader;
382 char *compressed_data = NULL;
384 size_t buffer_len = 0;
387 memset(&dkey, 0, sizeof(DBT));
388 memset(&ddata, 0, sizeof(DBT));
390 dkey.data = (void *) ckey;
391 ddata.size = cdatalen;
394 // "visit" records are numerous and have big, mostly-empty string buffers in them.
395 // If we compress these we can get them down to 1% of their size most of the time.
396 if (cdb == CDB_VISIT) {
398 zheader.magic = COMPRESS_MAGIC;
399 zheader.uncompressed_len = cdatalen;
400 buffer_len = ((cdatalen * 101) / 100) + 100 + sizeof(struct CtdlCompressHeader);
401 destLen = (uLongf) buffer_len;
402 compressed_data = malloc(buffer_len);
403 if (compress2((Bytef *) (compressed_data + sizeof(struct CtdlCompressHeader)), &destLen, (Bytef *) cdata, (uLongf) cdatalen, 1) != Z_OK) {
404 syslog(LOG_ERR, "bdb: compress2() error");
407 zheader.compressed_len = (size_t) destLen;
408 memcpy(compressed_data, &zheader, sizeof(struct CtdlCompressHeader));
409 ddata.size = (size_t) (sizeof(struct CtdlCompressHeader) + zheader.compressed_len);
410 ddata.data = compressed_data;
413 if (TSD->tid != NULL) {
414 ret = bdb_table[cdb]->put(bdb_table[cdb], // db
415 TSD->tid, // transaction ID
421 syslog(LOG_ERR, "bdb: bdb_store(%d): %s", cdb, db_strerror(ret));
425 free(compressed_data);
430 bdb_bailIfCursor(TSD->cursors, "attempt to write during r/o cursor");
435 if ((ret = bdb_table[cdb]->put(bdb_table[cdb], // db
436 tid, // transaction ID
440 if (ret == DB_LOCK_DEADLOCK) {
445 syslog(LOG_ERR, "bdb: bdb_store(%d): %s", cdb, db_strerror(ret));
452 free(compressed_data);
461 // Delete a piece of data. Returns 0 if the operation was successful.
462 int bdb_delete(int cdb, void *key, int keylen) {
467 memset(&dkey, 0, sizeof dkey);
471 if (TSD->tid != NULL) {
472 ret = bdb_table[cdb]->del(bdb_table[cdb], TSD->tid, &dkey, 0);
474 syslog(LOG_ERR, "bdb: bdb_delete(%d): %s", cdb, db_strerror(ret));
475 if (ret != DB_NOTFOUND) {
481 bdb_bailIfCursor(TSD->cursors, "attempt to delete during r/o cursor");
486 if ((ret = bdb_table[cdb]->del(bdb_table[cdb], tid, &dkey, 0)) && ret != DB_NOTFOUND) {
487 if (ret == DB_LOCK_DEADLOCK) {
492 syslog(LOG_ERR, "bdb: bdb_delete(%d): %s", cdb, db_strerror(ret));
504 static DBC *bdb_localcursor(int cdb) {
508 if (TSD->cursors[cdb] == NULL) {
509 ret = bdb_table[cdb]->cursor(bdb_table[cdb], TSD->tid, &curs, 0);
512 ret = TSD->cursors[cdb]->c_dup(TSD->cursors[cdb], &curs, DB_POSITION);
516 syslog(LOG_ERR, "bdb: bdb_localcursor: %s", db_strerror(ret));
524 // Fetch a piece of data. If not found, returns NULL. Otherwise, it returns
525 // a struct cdbdata which it is the caller's responsibility to free later on
526 // using the bdb_free() routine.
527 struct cdbdata *bdb_fetch(int cdb, const void *key, int keylen) {
529 if (keylen == 0) { // key length zero is impossible
533 struct cdbdata *tempcdb;
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));
565 tempcdb = (struct cdbdata *) malloc(sizeof(struct cdbdata));
566 if (tempcdb == NULL) {
567 syslog(LOG_ERR, "bdb: bdb_fetch() cannot allocate memory for tempcdb: %m");
571 tempcdb->len = TSD->dbdata[cdb].size;
572 tempcdb->ptr = TSD->dbdata[cdb].data;
573 bdb_decompress_if_necessary(tempcdb);
579 // Free a cdbdata item.
580 void bdb_free(struct cdbdata *cdb) {
581 if (cdb->cdbfree_must_free_ptr) {
588 void bdb_close_cursor(int cdb) {
589 if (TSD->cursors[cdb] != NULL) {
590 bdb_cclose(TSD->cursors[cdb]);
593 TSD->cursors[cdb] = NULL;
597 // Prepare for a sequential search of an entire database.
598 // (There is guaranteed to be no more than one traversal in progress per thread at any given time.)
599 void bdb_rewind(int cdb) {
602 if (TSD->cursors[cdb] != NULL) {
603 syslog(LOG_ERR, "bdb: bdb_rewind: must close cursor on database %d before reopening", cdb);
605 // bdb_cclose(TSD->cursors[cdb]);
608 // Now initialize the cursor
609 ret = bdb_table[cdb]->cursor(bdb_table[cdb], TSD->tid, &TSD->cursors[cdb], 0);
611 syslog(LOG_ERR, "bdb: bdb_rewind: db_cursor: %s", db_strerror(ret));
617 // Fetch the next item in a sequential search. Returns a pointer to a
618 // cdbdata structure, or NULL if we've hit the end.
619 struct cdbdata *bdb_next_item(int cdb) {
620 struct cdbdata *cdbret;
623 // reuse memory from the previous call.
624 TSD->dbkey[cdb].flags = DB_DBT_MALLOC;
625 TSD->dbdata[cdb].flags = DB_DBT_MALLOC;
627 ret = TSD->cursors[cdb]->c_get(TSD->cursors[cdb], &TSD->dbkey[cdb], &TSD->dbdata[cdb], DB_NEXT);
630 if (ret != DB_NOTFOUND) {
631 syslog(LOG_ERR, "bdb: bdb_next_item(%d): %s", cdb, db_strerror(ret));
634 bdb_close_cursor(cdb);
635 return NULL; // presumably, we are at the end
638 cdbret = (struct cdbdata *) malloc(sizeof(struct cdbdata));
639 cdbret->len = TSD->dbdata[cdb].size;
640 cdbret->ptr = TSD->dbdata[cdb].data;
641 bdb_decompress_if_necessary(cdbret);
647 // Transaction-based stuff. I'm writing this as I bake cookies...
648 void bdb_begin_transaction(void) {
649 bdb_bailIfCursor(TSD->cursors, "can't begin transaction during r/o cursor");
651 if (TSD->tid != NULL) {
652 syslog(LOG_ERR, "bdb: bdb_begin_transaction: ERROR: nested transaction");
656 bdb_txbegin(&TSD->tid);
660 void bdb_end_transaction(void) {
663 for (i = 0; i < MAXCDB; i++) {
664 if (TSD->cursors[i] != NULL) {
665 syslog(LOG_WARNING, "bdb: bdb_end_transaction: WARNING: cursor %d still open at transaction end", i);
666 bdb_cclose(TSD->cursors[i]);
667 TSD->cursors[i] = NULL;
671 if (TSD->tid == NULL) {
672 syslog(LOG_ERR, "bdb: bdb_end_transaction: ERROR: bdb_txcommit(NULL) !!");
676 bdb_txcommit(TSD->tid);
683 // Truncate (delete every record)
684 void bdb_trunc(int cdb) {
688 if (TSD->tid != NULL) {
689 syslog(LOG_ERR, "bdb: bdb_trunc must not be called in a transaction.");
693 bdb_bailIfCursor(TSD->cursors, "attempt to write during r/o cursor");
697 if ((ret = bdb_table[cdb]->truncate(bdb_table[cdb], // db
698 NULL, // transaction ID
699 &count, // #rows deleted
701 if (ret == DB_LOCK_DEADLOCK) {
705 syslog(LOG_ERR, "bdb: bdb_truncate(%d): %s", cdb, db_strerror(ret));
707 syslog(LOG_ERR, "bdb: You may need to tune your database; please read http://www.citadel.org for more information.");
716 // compact (defragment) the database, possibly returning space back to the underlying filesystem
717 void bdb_compact(void) {
721 syslog(LOG_DEBUG, "bdb: bdb_compact() started");
722 for (i = 0; i < MAXCDB; i++) {
723 syslog(LOG_DEBUG, "bdb: compacting database %d", i);
724 ret = bdb_table[i]->compact(bdb_table[i], NULL, NULL, NULL, NULL, DB_FREE_SPACE, NULL);
726 syslog(LOG_ERR, "bdb: compact: %s", db_strerror(ret));
729 syslog(LOG_DEBUG, "bdb: bdb_compact() finished");
733 // Calling this function activates the Berkeley DB back end.
734 void bdb_init_backend(void) {
736 // Assign the backend API stubs to the functions in this module.
737 cdb_compact = bdb_compact;
738 cdb_checkpoint = bdb_checkpoint;
739 cdb_rewind = bdb_rewind;
740 cdb_fetch = bdb_fetch;
741 cdb_open_databases = bdb_open_databases;
742 cdb_close_databases = bdb_close_databases;
743 cdb_store = bdb_store;
744 cdb_delete = bdb_delete;
746 cdb_next_item = bdb_next_item;
747 cdb_close_cursor = bdb_close_cursor;
748 cdb_begin_transaction = bdb_begin_transaction;
749 cdb_end_transaction = bdb_end_transaction;
750 cdb_check_handles = bdb_check_handles;
751 cdb_trunc = bdb_trunc;
753 // Some functions in this backend need to store some per-thread data.
754 // We crerate the key here, during module initialization.
755 if (pthread_key_create(&bdb_thread_key, NULL) != 0) {
756 syslog(LOG_ERR, "pthread_key_create() : %m");
757 exit(CTDLEXIT_THREAD);
760 syslog(LOG_INFO, "db: initialized Berkeley DB backend");