/*
- * $Id: main.c,v 1.2 2005-04-28 20:49:48 bfernhomberg Exp $
- *
* Copyright (C) Joerg Lenneis 2003
+ * Copyright (c) Frank Lahm 2009
* All Rights Reserved. See COPYING.
*/
#include <netatalk/endian.h>
#include <atalk/cnid_dbd_private.h>
#include <atalk/logger.h>
+#include <atalk/volinfo.h>
#include "db_param.h"
#include "dbif.h"
#include "dbd.h"
#include "comm.h"
+/*
+ Note: DB_INIT_LOCK is here so we can run the db_* utilities while netatalk is running.
+ It's a likey performance hit, but it might we worth it.
+ */
+#define DBOPTIONS (DB_CREATE | DB_INIT_LOG | DB_INIT_MPOOL | DB_INIT_LOCK | DB_INIT_TXN)
-#define LOCKFILENAME "lock"
+/* Global, needed by pack.c:idxname() */
+struct volinfo volinfo;
+static DBD *dbd;
static int exit_sig = 0;
-
+static int db_locked;
static void sig_exit(int signo)
{
static void block_sigs_onoff(int block)
{
sigset_t set;
-
+
sigemptyset(&set);
sigaddset(&set, SIGINT);
sigaddset(&set, SIGTERM);
return;
}
-/*
- The dbd_XXX and comm_XXX functions all obey the same protocol for return values:
-
- 1: Success, if transactions are used commit.
- 0: Failure, but we continue to serve requests. If transactions are used abort/rollback.
- -1: Fatal error, either from the database or from the socket. Abort the transaction if applicable
- (which might fail as well) and then exit.
+/*
+ The dbd_XXX and comm_XXX functions all obey the same protocol for return values:
+
+ 1: Success, if transactions are used commit.
+ 0: Failure, but we continue to serve requests. If transactions are used abort/rollback.
+ -1: Fatal error, either from t
+ he database or from the socket. Abort the transaction if applicable
+ (which might fail as well) and then exit.
We always try to notify the client process about the outcome, the result field
of the cnid_dbd_rply structure contains further details.
- */
+*/
+#ifndef min
+#define min(a,b) ((a)<(b)?(a):(b))
+#endif
static int loop(struct db_param *dbp)
{
struct cnid_dbd_rqst rqst;
struct cnid_dbd_rply rply;
+ time_t timeout;
int ret, cret;
- time_t now, time_next_flush, time_last_rqst;
int count;
+ time_t now, time_next_flush, time_last_rqst;
+ char timebuf[64];
static char namebuf[MAXPATHLEN + 1];
- u_int32_t checkp_flags;
+ sigset_t set;
+
+ sigemptyset(&set);
+ sigprocmask(SIG_SETMASK, NULL, &set);
+ sigdelset(&set, SIGINT);
+ sigdelset(&set, SIGTERM);
count = 0;
now = time(NULL);
time_next_flush = now + dbp->flush_interval;
time_last_rqst = now;
- if (dbp->nosync)
- checkp_flags = DB_FORCE;
- else
- checkp_flags = 0;
-
+
rqst.name = namebuf;
+ strftime(timebuf, 63, "%b %d %H:%M:%S.",localtime(&time_next_flush));
+ LOG(log_debug, logtype_cnid, "Checkpoint interval: %d seconds. Next checkpoint: %s",
+ dbp->flush_interval, timebuf);
+
while (1) {
- if ((cret = comm_rcv(&rqst)) < 0)
+ timeout = min(time_next_flush, time_last_rqst +dbp->idle_timeout);
+ if (timeout > now)
+ timeout -= now;
+ else
+ timeout = 1;
+
+ if ((cret = comm_rcv(&rqst, timeout, &set, &now)) < 0)
return -1;
- now = time(NULL);
-
- if (count > dbp->flush_frequency || now > time_next_flush) {
-#ifdef CNID_BACKEND_DBD_TXN
- if (dbif_txn_checkpoint(0, 0, checkp_flags) < 0)
+ if (cret == 0) {
+ /* comm_rcv returned from select without receiving anything. */
+ if (exit_sig) {
+ /* Received signal (TERM|INT) */
+ return 0;
+ }
+ if (now - time_last_rqst >= dbp->idle_timeout && comm_nbe() <= 0) {
+ /* Idle timeout */
+ return 0;
+ }
+ /* still active connections, reset time_last_rqst */
+ time_last_rqst = now;
+ } else {
+ /* We got a request */
+ time_last_rqst = now;
+
+ memset(&rply, 0, sizeof(rply));
+ switch(rqst.op) {
+ /* ret gets set here */
+ case CNID_DBD_OP_OPEN:
+ case CNID_DBD_OP_CLOSE:
+ /* open/close are noops for now. */
+ rply.namelen = 0;
+ ret = 1;
+ break;
+ case CNID_DBD_OP_ADD:
+ ret = dbd_add(dbd, &rqst, &rply, 0);
+ break;
+ case CNID_DBD_OP_GET:
+ ret = dbd_get(dbd, &rqst, &rply);
+ break;
+ case CNID_DBD_OP_RESOLVE:
+ ret = dbd_resolve(dbd, &rqst, &rply);
+ break;
+ case CNID_DBD_OP_LOOKUP:
+ ret = dbd_lookup(dbd, &rqst, &rply, 0);
+ break;
+ case CNID_DBD_OP_UPDATE:
+ ret = dbd_update(dbd, &rqst, &rply);
+ break;
+ case CNID_DBD_OP_DELETE:
+ ret = dbd_delete(dbd, &rqst, &rply, DBIF_CNID);
+ break;
+ case CNID_DBD_OP_GETSTAMP:
+ ret = dbd_getstamp(dbd, &rqst, &rply);
+ break;
+ case CNID_DBD_OP_REBUILD_ADD:
+ ret = dbd_rebuild_add(dbd, &rqst, &rply);
+ break;
+ case CNID_DBD_OP_SEARCH:
+ ret = dbd_search(dbd, &rqst, &rply);
+ break;
+ default:
+ LOG(log_error, logtype_cnid, "loop: unknown op %d", rqst.op);
+ ret = -1;
+ break;
+ }
+
+ if ((cret = comm_snd(&rply)) < 0 || ret < 0) {
+ dbif_txn_abort(dbd);
return -1;
-#else
- if (dbif_sync() < 0)
+ }
+
+ if (ret == 0 || cret == 0) {
+ if (dbif_txn_abort(dbd) < 0)
+ return -1;
+ } else {
+ ret = dbif_txn_commit(dbd);
+ if ( ret < 0)
+ return -1;
+ else if ( ret > 0 )
+ /* We had a designated txn because we wrote to the db */
+ count++;
+ }
+ } /* got a request */
+
+ /*
+ Shall we checkpoint bdb ?
+ "flush_interval" seconds passed ?
+ */
+ if (now >= time_next_flush) {
+ LOG(log_info, logtype_cnid, "Checkpointing BerkeleyDB for volume '%s'", dbp->dir);
+ if (dbif_txn_checkpoint(dbd, 0, 0, 0) < 0)
return -1;
-#endif
count = 0;
time_next_flush = now + dbp->flush_interval;
- }
- if (cret == 0) {
- block_sigs_onoff(0);
- block_sigs_onoff(1);
- if (exit_sig)
- return 0;
- if (dbp->idle_timeout && comm_nbe() <= 0 && (now - time_last_rqst) > dbp->idle_timeout)
- return 0;
- continue;
+ strftime(timebuf, 63, "%b %d %H:%M:%S.",localtime(&time_next_flush));
+ LOG(log_debug, logtype_cnid, "Checkpoint interval: %d seconds. Next checkpoint: %s",
+ dbp->flush_interval, timebuf);
}
- /* We got a request */
- time_last_rqst = now;
- count++;
-
-#ifdef CNID_BACKEND_DBD_TXN
- if (dbif_txn_begin() < 0)
- return -1;
-#endif /* CNID_BACKEND_DBD_TXN */
-
- memset(&rply, 0, sizeof(rply));
- switch(rqst.op) {
- /* ret gets set here */
- case CNID_DBD_OP_OPEN:
- case CNID_DBD_OP_CLOSE:
- /* open/close are noops for now. */
- rply.namelen = 0;
- ret = 1;
- break;
- case CNID_DBD_OP_ADD:
- ret = dbd_add(&rqst, &rply);
- break;
- case CNID_DBD_OP_GET:
- ret = dbd_get(&rqst, &rply);
- break;
- case CNID_DBD_OP_RESOLVE:
- ret = dbd_resolve(&rqst, &rply);
- break;
- case CNID_DBD_OP_LOOKUP:
- ret = dbd_lookup(&rqst, &rply);
- break;
- case CNID_DBD_OP_UPDATE:
- ret = dbd_update(&rqst, &rply);
- break;
- case CNID_DBD_OP_DELETE:
- ret = dbd_delete(&rqst, &rply);
- break;
- case CNID_DBD_OP_GETSTAMP:
- ret = dbd_getstamp(&rqst, &rply);
- break;
- case CNID_DBD_OP_REBUILD_ADD:
- ret = dbd_rebuild_add(&rqst, &rply);
- break;
- default:
- LOG(log_error, logtype_cnid, "loop: unknown op %d", rqst.op);
- ret = -1;
- break;
- }
- if ((cret = comm_snd(&rply)) < 0 || ret < 0) {
-#ifdef CNID_BACKEND_DBD_TXN
- dbif_txn_abort();
-#endif /* CNID_BACKEND_DBD_TXN */
- return -1;
- }
-#ifdef CNID_BACKEND_DBD_TXN
- if (ret == 0 || cret == 0) {
- if (dbif_txn_abort() < 0)
- return -1;
- } else {
- if (dbif_txn_commit() < 0)
+ /*
+ Shall we checkpoint bdb ?
+ Have we commited "count" more changes than "flush_frequency" ?
+ */
+ if (count > dbp->flush_frequency) {
+ LOG(log_info, logtype_cnid, "Checkpointing BerkeleyDB after %d writes for volume '%s'", count, dbp->dir);
+ if (dbif_txn_checkpoint(dbd, 0, 0, 0) < 0)
return -1;
+ count = 0;
}
-#endif /* CNID_BACKEND_DBD_TXN */
- }
+ } /* while(1) */
}
/* ------------------------ */
LOG(log_error, logtype_cnid, "chdir to %s failed: %s", dir, strerror(errno));
exit(1);
}
-
+
if (stat(".", &st) < 0) {
LOG(log_error, logtype_cnid, "error in stat for %s: %s", dir, strerror(errno));
exit(1);
}
}
-/* ------------------------ */
-int get_lock(void)
-{
- int lockfd;
- struct flock lock;
-
- if ((lockfd = open(LOCKFILENAME, O_RDWR | O_CREAT, 0644)) < 0) {
- LOG(log_error, logtype_cnid, "main: error opening lockfile: %s", strerror(errno));
- exit(1);
- }
-
- lock.l_start = 0;
- lock.l_whence = SEEK_SET;
- lock.l_len = 0;
- lock.l_type = F_WRLCK;
-
- if (fcntl(lockfd, F_SETLK, &lock) < 0) {
- if (errno == EACCES || errno == EAGAIN) {
- exit(0);
- } else {
- LOG(log_error, logtype_cnid, "main: fcntl F_WRLCK lockfile: %s", strerror(errno));
- exit(1);
- }
- }
-
- return lockfd;
-}
/* ----------------------- */
-void set_signal(void)
+static void set_signal(void)
{
struct sigaction sv;
}
}
-/* ----------------------- */
-void free_lock(int lockfd)
-{
- struct flock lock;
-
- lock.l_start = 0;
- lock.l_whence = SEEK_SET;
- lock.l_len = 0;
- lock.l_type = F_UNLCK;
- fcntl(lockfd, F_SETLK, &lock);
- close(lockfd);
-}
-
/* ------------------------ */
int main(int argc, char *argv[])
{
struct db_param *dbp;
- int err = 0;
- int ret;
- int lockfd, ctrlfd, clntfd;
- char *dir;
-
+ int err = 0, ret, delete_bdb = 0;
+ int ctrlfd, clntfd;
+ char *logconfig;
+
set_processname("cnid_dbd");
- syslog_setup(log_debug, logtype_default, logoption_ndelay | logoption_pid, logfacility_daemon);
-
- if (argc != 4) {
+
+ while (( ret = getopt( argc, argv, "vVd")) != -1 ) {
+ switch (ret) {
+ case 'v':
+ case 'V':
+ printf("cnid_dbd (Netatalk %s)\n", VERSION);
+ return -1;
+ case 'd':
+ delete_bdb = 1;
+ break;
+ }
+ }
+
+ if (argc - optind != 4) {
LOG(log_error, logtype_cnid, "main: not enough arguments");
+ exit(EXIT_FAILURE);
+ }
+
+ /* Load .volinfo file */
+ if (loadvolinfo(argv[optind], &volinfo) == -1) {
+ LOG(log_error, logtype_cnid, "Cant load volinfo for \"%s\"", argv[1]);
+ exit(EXIT_FAILURE);
+ }
+ /* Put "/.AppleDB" at end of volpath, get path from volinfo file */
+ char dbpath[MAXPATHLEN+1];
+ if ((strlen(volinfo.v_dbpath) + strlen("/.AppleDB")) > MAXPATHLEN ) {
+ LOG(log_error, logtype_cnid, "CNID db pathname too long: \"%s\"", volinfo.v_dbpath);
+ exit(EXIT_FAILURE);
+ }
+ strncpy(dbpath, volinfo.v_dbpath, MAXPATHLEN - strlen("/.AppleDB"));
+ strcat(dbpath, "/.AppleDB");
+
+ ctrlfd = atoi(argv[optind + 1]);
+ clntfd = atoi(argv[optind + 2]);
+ logconfig = strdup(argv[optind + 3]);
+ setuplog(logconfig);
+
+ if (vol_load_charsets(&volinfo) == -1) {
+ LOG(log_error, logtype_cnid, "Error loading charsets!");
+ exit(EXIT_FAILURE);
+ }
+ LOG(log_debug, logtype_cnid, "db dir: \"%s\"", dbpath);
+
+ switch_to_user(dbpath);
+
+ /* Get db lock */
+ if ((db_locked = get_lock(LOCK_EXCL, dbpath)) == -1) {
+ LOG(log_error, logtype_cnid, "main: fatal db lock error");
exit(1);
}
+ if (db_locked != LOCK_EXCL) {
+ /* Couldn't get exclusive lock, try shared lock */
+ if ((db_locked = get_lock(LOCK_SHRD, NULL)) != LOCK_SHRD) {
+ LOG(log_error, logtype_cnid, "main: fatal db lock error");
+ exit(1);
+ }
+ }
+
+ if (delete_bdb && (db_locked == LOCK_EXCL)) {
+ LOG(log_warning, logtype_cnid, "main: too many CNID db opening attempts, wiping the slate clean");
+ chdir(dbpath);
+ system("rm -f cnid2.db lock log.* __db.*");
+ if ((db_locked = get_lock(LOCK_EXCL, dbpath)) != LOCK_EXCL) {
+ LOG(log_error, logtype_cnid, "main: fatal db lock error");
+ exit(EXIT_FAILURE);
+ }
+ }
- dir = argv[1];
- ctrlfd = atoi(argv[2]);
- clntfd = atoi(argv[3]);
-
- switch_to_user(dir);
-
- /* Before we do anything else, check if there is an instance of cnid_dbd
- running already and silently exit if yes. */
- lockfd = get_lock();
-
- LOG(log_info, logtype_cnid, "Startup, DB dir %s", dir);
-
set_signal();
-
- /* SIGINT and SIGTERM are always off, unless we get a return code of 0 from
- comm_rcv (no requests for one second, see above in loop()). That means we
- only shut down after one second of inactivity. */
+
+ /* SIGINT and SIGTERM are always off, unless we are in pselect */
block_sigs_onoff(1);
- if ((dbp = db_param_read(dir)) == NULL)
+ if ((dbp = db_param_read(dbpath)) == NULL)
exit(1);
+ LOG(log_maxdebug, logtype_cnid, "Finished parsing db_param config file");
- if (dbif_env_init(dbp) < 0)
- exit(2); /* FIXME: same exit code as failure for dbif_open() */
-
-#ifdef CNID_BACKEND_DBD_TXN
- if (dbif_txn_begin() < 0)
- exit(6);
-#endif
-
- if (dbif_open(dbp, 0) < 0) {
-#ifdef CNID_BACKEND_DBD_TXN
- dbif_txn_abort();
-#endif
- dbif_close();
+ if (NULL == (dbd = dbif_init(dbpath, "cnid2.db")))
+ exit(2);
+
+ /* Only recover if we got the lock */
+ if (dbif_env_open(dbd,
+ dbp,
+ (db_locked == LOCK_EXCL) ? DBOPTIONS | DB_RECOVER : DBOPTIONS) < 0)
+ exit(2); /* FIXME: same exit code as failure for dbif_open() */
+ LOG(log_debug, logtype_cnid, "Finished initializing BerkeleyDB environment");
+
+ if (dbif_open(dbd, dbp, 0) < 0) {
+ dbif_close(dbd);
exit(2);
}
+ LOG(log_debug, logtype_cnid, "Finished opening BerkeleyDB databases");
-#ifndef CNID_BACKEND_DBD_TXN
- if (dbp->check && (ret = dbd_check(dir))) {
- if (ret < 0) {
- dbif_close();
+ /* Downgrade db lock */
+ if (db_locked == LOCK_EXCL) {
+ if (get_lock(LOCK_UNLOCK, NULL) != 0) {
+ dbif_close(dbd);
exit(2);
}
- dbif_closedb();
- LOG(log_info, logtype_cnid, "main: re-opening, secondaries will be rebuilt. This may take some time");
- if (dbif_open(dbp, 1) < 0) {
- LOG(log_info, logtype_cnid, "main: re-opening databases failed");
- dbif_close();
+ if (get_lock(LOCK_SHRD, NULL) != LOCK_SHRD) {
+ dbif_close(dbd);
exit(2);
}
- LOG(log_info, logtype_cnid, "main: rebuilt done");
}
-#endif
- if (dbd_stamp() < 0) {
-#ifdef CNID_BACKEND_DBD_TXN
- dbif_txn_abort();
-#endif
- dbif_close();
- exit(5);
- }
-#ifdef CNID_BACKEND_DBD_TXN
- if (dbif_txn_commit() < 0)
- exit(6);
-#endif
if (comm_init(dbp, ctrlfd, clntfd) < 0) {
- dbif_close();
+ dbif_close(dbd);
exit(3);
}
if (loop(dbp) < 0)
err++;
-#ifndef CNID_BACKEND_DBD_TXN
- /* FIXME: Do we really need to sync before closing the DB? Just closing it
- should be enough. */
- if (dbif_sync() < 0)
+ if (dbif_close(dbd) < 0)
err++;
-#endif
- if (dbif_close() < 0)
+ if (dbif_env_remove(dbpath) < 0)
err++;
-
- free_lock(lockfd);
-
+
if (err)
exit(4);
else if (exit_sig)
- LOG(log_info, logtype_cnid, "main: Exiting on signal %i", exit_sig);
+ LOG(log_info, logtype_cnid, "main: Exiting on signal %i", exit_sig);
else
- LOG(log_info, logtype_cnid, "main: Idle timeout, exiting");
-
+ LOG(log_info, logtype_cnid, "main: Idle timeout, exiting");
+
return 0;
}