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 // Panic callback for Berkeley DB. Not part of the backend API.
116 static void bdb_dbpanic(DB_ENV *env, int errval) {
117 syslog(LOG_ERR, "bdb: PANIC: %s", db_strerror(errval));
122 // Close a cursor -- not part of the backend API.
123 static void bdb_cclose(DBC *cursor) {
126 if ((ret = cursor->c_close(cursor))) {
127 syslog(LOG_ERR, "bdb: c_close: %s", db_strerror(ret));
133 // Convenience function to abort if a cursor is still open when we didn't expect it to be.
134 // This is not part of the backend API.
135 static void bdb_bailIfCursor(DBC **cursors, const char *msg) {
138 for (i = 0; i < MAXCDB; i++)
139 if (cursors[i] != NULL) {
140 syslog(LOG_ERR, "bdb: cursor still in progress on cdb %02x: %s", i, msg);
146 void bdb_check_handles(void) {
147 bdb_bailIfCursor(TSD->cursors, "in check_handles");
149 if (TSD->tid != NULL) {
150 syslog(LOG_ERR, "bdb: transaction still in progress!");
156 // Transaction-based stuff. I'm writing this as I bake cookies...
157 void bdb_begin_transaction(void) {
159 bdb_bailIfCursor(TSD->cursors, "can't begin transaction during r/o cursor");
161 if (TSD->tid != NULL) {
162 syslog(LOG_ERR, "bdb: bdb_begin_transaction: ERROR: nested transaction");
166 ret = bdb_env->txn_begin(bdb_env, NULL, &TSD->tid, 0);
168 syslog(LOG_ERR, "bdb: bdb_begin_transaction: %s", db_strerror(ret));
174 // ...and the cookies are cursed.
175 void bdb_end_transaction(void) {
178 for (i = 0; i < MAXCDB; i++) {
179 if (TSD->cursors[i] != NULL) {
180 syslog(LOG_WARNING, "bdb: bdb_end_transaction: WARNING: cursor %d still open at transaction end", i);
181 bdb_cclose(TSD->cursors[i]);
182 TSD->cursors[i] = NULL;
186 if (TSD->tid == NULL) {
187 syslog(LOG_ERR, "bdb: bdb_end_transaction: ERROR: bdb_txcommit(NULL) !!");
191 bdb_txcommit(TSD->tid);
198 // Request a checkpoint of the database. Called once per minute by the thread manager.
199 void bdb_checkpoint(void) {
202 syslog(LOG_DEBUG, "bdb: -- checkpoint --");
203 ret = bdb_env->txn_checkpoint(bdb_env, MAX_CHECKPOINT_KBYTES, MAX_CHECKPOINT_MINUTES, 0);
206 syslog(LOG_ERR, "bdb: bdb_checkpoint() txn_checkpoint: %s", db_strerror(ret));
210 // After a successful checkpoint, we can cull the unused logs
211 ret = bdb_env->log_set_config(bdb_env, DB_LOG_AUTO_REMOVE, 1);
213 syslog(LOG_ERR, "bdb: bdb_checkpoint() auto coll logs: %s", db_strerror(ret));
218 // Open the various tables we'll be using. Any table which
219 // does not exist should be created. Note that we don't need a
220 // critical section here, because there aren't any active threads
221 // manipulating the database yet.
222 void bdb_open_databases(void) {
227 int dbversion_major, dbversion_minor, dbversion_patch;
229 syslog(LOG_DEBUG, "bdb: bdb_open_databases() starting");
230 syslog(LOG_DEBUG, "bdb: Linked zlib: %s", zlibVersion());
231 syslog(LOG_DEBUG, "bdb: Compiled libdb: %s", DB_VERSION_STRING);
232 syslog(LOG_DEBUG, "bdb: Linked libdb: %s", db_version(&dbversion_major, &dbversion_minor, &dbversion_patch));
234 // Create synthetic integer version numbers and compare them.
235 // Never allow citserver to run with a libdb older then the one with which it was compiled.
236 int compiled_db_version = ( (DB_VERSION_MAJOR * 1000000) + (DB_VERSION_MINOR * 1000) + (DB_VERSION_PATCH) );
237 int linked_db_version = ( (dbversion_major * 1000000) + (dbversion_minor * 1000) + (dbversion_patch) );
238 if (compiled_db_version > linked_db_version) {
239 syslog(LOG_ERR, "bdb: citserver is running with a version of libdb older than the one with which it was compiled.");
240 syslog(LOG_ERR, "bdb: This is an invalid configuration. citserver will now exit to prevent data loss.");
244 syslog(LOG_DEBUG, "bdb: Setting up DB environment");
245 ret = db_env_create(&bdb_env, 0);
247 syslog(LOG_ERR, "bdb: db_env_create: %s", db_strerror(ret));
248 syslog(LOG_ERR, "bdb: exit code %d", ret);
251 bdb_env->set_errpfx(bdb_env, "citserver");
252 bdb_env->set_paniccall(bdb_env, bdb_dbpanic);
253 bdb_env->set_errcall(bdb_env, bdb_verbose_err);
254 bdb_env->set_msgcall(bdb_env, bdb_verbose_log);
255 bdb_env->set_verbose(bdb_env, DB_VERB_DEADLOCK, 1);
256 bdb_env->set_verbose(bdb_env, DB_VERB_RECOVERY, 1);
258 flags = DB_CREATE | DB_INIT_LOCK | DB_INIT_LOG | DB_INIT_MPOOL | DB_INIT_TXN | DB_RECOVER | DB_THREAD ;
259 syslog(LOG_DEBUG, "bdb: bdb_env->open(bdb_env, %s, %d, 0)", ctdl_db_dir, flags);
260 ret = bdb_env->open(bdb_env, ctdl_db_dir, flags, 0); // try opening the database cleanly
262 syslog(LOG_ERR, "bdb: bdb_env->open: %s: %s", ctdl_db_dir, db_strerror(ret));
263 bdb_env->close(bdb_env, 0);
264 syslog(LOG_ERR, "bdb: exit code %d", ret);
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, 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 ret = bdb_table[i]->close(bdb_table[i], 0);
316 syslog(LOG_ERR, "bdb: db_close: %s", db_strerror(ret));
321 syslog(LOG_INFO, "bdb: closing environment");
322 ret = bdb_env->close(bdb_env, DB_FORCESYNC);
324 syslog(LOG_ERR, "bdb: DBENV->close: %s", db_strerror(ret));
327 syslog(LOG_INFO, "bdb: shutdown completed");
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;
385 int existing_txn = 0; // set to nonzero if we are already inside a transaction
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) {
417 if (!existing_txn) { // If we're not already inside a transaction,
418 bdb_begin_transaction(); // create our own for this operation.
422 ret = bdb_table[cdb]->put(bdb_table[cdb], // db
423 TSD->tid, // transaction ID
428 if ((ret != 0) && (ret != DB_LOCK_DEADLOCK)) {
429 syslog(LOG_ERR, "bdb: bdb_store(%02x): error %d: %s", cdb, ret, db_strerror(ret));
432 if (ret == DB_LOCK_DEADLOCK) {
433 syslog(LOG_DEBUG, "bdb: bdb_store(%02x): would deadlock, trying again", cdb);
435 } while (ret == DB_LOCK_DEADLOCK);
438 bdb_end_transaction();
442 free(compressed_data);
449 // Delete a piece of data. Returns 0 if the operation was successful.
450 int bdb_delete(int cdb, void *key, int keylen) {
453 int existing_txn = 0; // set to nonzero if we are already inside a transaction
455 memset(&dkey, 0, sizeof dkey);
459 if (TSD->tid != NULL) {
463 if (!existing_txn) { // If we're not already inside a transaction,
464 bdb_begin_transaction(); // create our own for this operation.
467 ret = bdb_table[cdb]->del(bdb_table[cdb], TSD->tid, &dkey, 0);
469 if (ret != DB_NOTFOUND) {
470 syslog(LOG_ERR, "bdb: bdb_delete(%02x): %s", cdb, db_strerror(ret));
476 bdb_end_transaction(); // Only end the transaction if we began it.
483 // Fetch a piece of data. Returns a "struct cdbdata"
484 // If the item is not found, the pointer will be NULL.
485 struct cdbdata bdb_fetch(int cdb, const void *key, int keylen) {
487 struct cdbdata returned_data;
488 memset(&returned_data, 0, sizeof(struct cdbdata));
490 if (keylen == 0) { // key length zero is impossible
491 return(returned_data);
497 memset(&dkey, 0, sizeof(DBT));
499 dkey.data = (void *) key;
501 TSD->dbdata[cdb].flags = DB_DBT_REALLOC;
504 ret = bdb_table[cdb]->get(bdb_table[cdb], TSD->tid, &dkey, &TSD->dbdata[cdb], 0);
505 if ((ret != 0) && (ret != DB_NOTFOUND) && (ret != DB_LOCK_DEADLOCK)) {
506 syslog(LOG_ERR, "bdb: bdb_fetch(%d): error %d: %s", cdb, ret, db_strerror(ret));
509 } while (ret == DB_LOCK_DEADLOCK);
512 bdb_decompress_if_necessary(&TSD->dbdata[cdb]);
513 returned_data.len = TSD->dbdata[cdb].size;
514 returned_data.ptr = TSD->dbdata[cdb].data;
517 return(returned_data);
521 void bdb_close_cursor(int cdb) {
522 if (TSD->cursors[cdb] != NULL) {
523 bdb_cclose(TSD->cursors[cdb]);
526 TSD->cursors[cdb] = NULL;
530 // Prepare for a sequential search of an entire database.
531 // (There is guaranteed to be no more than one traversal in progress per thread at any given time.)
532 void bdb_rewind(int cdb) {
535 if (TSD->cursors[cdb] != NULL) {
536 syslog(LOG_ERR, "bdb: bdb_rewind: must close cursor on database %d before reopening", cdb);
538 // bdb_cclose(TSD->cursors[cdb]);
541 // Now initialize the cursor
542 ret = bdb_table[cdb]->cursor(bdb_table[cdb], TSD->tid, &TSD->cursors[cdb], 0);
544 syslog(LOG_ERR, "bdb: bdb_rewind: db_cursor: %s", db_strerror(ret));
550 // Fetch the next item in a sequential search. Returns a pointer to a
551 // cdbdata structure, or NULL if we've hit the end.
552 struct cdbkeyval bdb_next_item(int cdb) {
556 memset(&kv, 0, sizeof(struct cdbkeyval));
558 // reuse memory from the previous call.
559 TSD->dbkey[cdb].flags = DB_DBT_REALLOC;
560 TSD->dbdata[cdb].flags = DB_DBT_REALLOC;
562 assert(TSD->cursors[cdb] != NULL);
563 ret = TSD->cursors[cdb]->c_get(TSD->cursors[cdb], &TSD->dbkey[cdb], &TSD->dbdata[cdb], DB_NEXT);
566 if (ret != DB_NOTFOUND) {
567 syslog(LOG_ERR, "bdb: bdb_next_item(%d): %s", cdb, db_strerror(ret));
570 bdb_close_cursor(cdb);
571 return(kv); // presumably, we are at the end
574 bdb_decompress_if_necessary(&TSD->dbdata[cdb]);
576 kv.key.len = TSD->dbkey[cdb].size;
577 kv.key.ptr = TSD->dbkey[cdb].data;
578 kv.val.len = TSD->dbdata[cdb].size;
579 kv.val.ptr = TSD->dbdata[cdb].data;
584 // Truncate (delete every record)
585 void bdb_trunc(int cdb) {
589 if (TSD->tid != NULL) {
590 syslog(LOG_ERR, "bdb: bdb_trunc must not be called in a transaction.");
594 bdb_begin_transaction(); // create our own transaction for this operation.
595 ret = bdb_table[cdb]->truncate(bdb_table[cdb], // db
596 NULL, // transaction ID
597 &count, // #rows deleted
601 syslog(LOG_ERR, "bdb: bdb_truncate(%d): %s", cdb, db_strerror(ret));
603 syslog(LOG_ERR, "bdb: You may need to tune your database; please read http://www.citadel.org for more information.");
607 bdb_end_transaction();
611 // compact (defragment) the database, possibly returning space back to the underlying filesystem
612 void bdb_compact(void) {
616 syslog(LOG_DEBUG, "bdb: bdb_compact() started");
617 for (i = 0; i < MAXCDB; i++) {
618 syslog(LOG_DEBUG, "bdb: compacting database %d", i);
619 ret = bdb_table[i]->compact(bdb_table[i], NULL, NULL, NULL, NULL, DB_FREE_SPACE, NULL);
621 syslog(LOG_ERR, "bdb: compact: %s", db_strerror(ret));
624 syslog(LOG_DEBUG, "bdb: bdb_compact() finished");
628 // periodically called for maintenance
629 void bdb_tick(void) {
633 ret = bdb_env->lock_detect(bdb_env, 0, DB_LOCK_DEFAULT, &rejected);
635 syslog(LOG_ERR, "bdb: lock_detect: %s", db_strerror(ret));
638 syslog(LOG_DEBUG, "bdb: rejected lock %d", rejected);
642 // Calling this function activates the Berkeley DB back end.
643 void bdb_init_backend(void) {
645 // Assign the backend API stubs to the functions in this module.
646 cdb_compact = bdb_compact;
647 cdb_checkpoint = bdb_checkpoint;
648 cdb_rewind = bdb_rewind;
649 cdb_fetch = bdb_fetch;
650 cdb_open_databases = bdb_open_databases;
651 cdb_close_databases = bdb_close_databases;
652 cdb_store = bdb_store;
653 cdb_delete = bdb_delete;
654 cdb_next_item = bdb_next_item;
655 cdb_close_cursor = bdb_close_cursor;
656 cdb_begin_transaction = bdb_begin_transaction;
657 cdb_end_transaction = bdb_end_transaction;
658 cdb_check_handles = bdb_check_handles;
659 cdb_trunc = bdb_trunc;
662 // Some functions in this backend need to store some per-thread data.
663 // We crerate the key here, during module initialization.
664 if (pthread_key_create(&bdb_thread_key, NULL) != 0) {
665 syslog(LOG_ERR, "pthread_key_create() : %m");
666 exit(CTDLEXIT_THREAD);
669 syslog(LOG_INFO, "db: initialized Berkeley DB backend");