X-Git-Url: http://git.tuebingen.mpg.de/?p=dss.git;a=blobdiff_plain;f=dss.c;h=b9afbc4a8505d1886420d1355bd742dbdd73c756;hp=744efaf0965f5a44c68d53d43163ae8dcbb6b209;hb=442d1920cb866fb6e0a816d7e175b5cdb8c0e1a8;hpb=3de0718ad11e13e063e5a24be5758883c48c5ac2

diff --git a/dss.c b/dss.c
index 744efaf..b9afbc4 100644
--- a/dss.c
+++ b/dss.c
@@ -1,3 +1,8 @@
+/*
+ * Copyright (C) 2008 Andre Noll <maan@systemlinux.org>
+ *
+ * Licensed under the GPL v2. For licencing details see COPYING.
+ */
 #include <string.h>
 #include <stdlib.h>
 #include <stdarg.h>
@@ -35,24 +40,26 @@ static struct gengetopt_args_info conf;
 static FILE *logfile;
 /** The read end of the signal pipe */
 static int signal_pipe;
-/** Process id of current rsync process. */
-static pid_t rsync_pid;
-/** Whether the rsync process is currently stopped */
-static int rsync_stopped;
-/** Process id of current rm process. */
-static pid_t rm_pid;
+/** Process id of current pre-create-hook/rsync/post-create-hook process. */
+static pid_t create_pid;
+/** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
+static int create_process_stopped;
+/** Process id of current pre-remove/rm/post-remove process. */
+static pid_t remove_pid;
 /** When the next snapshot is due. */
 static struct timeval next_snapshot_time;
-/** The pid of the pre-create hook. */
-static pid_t pre_create_hook_pid;
-/** The pid of the post-create hook. */
-static pid_t post_create_hook_pid;
+/** When to try to remove something. */
+static struct timeval next_removal_check;
 /** Creation time of the snapshot currently being created. */
 static int64_t current_snapshot_creation_time;
+/* The snapshot currently being removed. */
+struct snapshot *snapshot_currently_being_removed;
 /** Needed by the post-create hook. */
 static char *path_to_last_complete_snapshot;
 /** \sa \ref snap.h for details. */
-static unsigned snapshot_creation_status;
+enum hook_status snapshot_creation_status;
+/** \sa \ref snap.h for details. */
+enum hook_status snapshot_removal_status;
 
 
 DEFINE_DSS_ERRLIST;
@@ -72,7 +79,7 @@ static int call_command_handler(void)
 {
 	COMMANDS
 	DSS_EMERG_LOG("BUG: did not find command handler\n");
-	exit(EXIT_FAILURE);
+	return -E_BUG;
 }
 #undef COMMAND
 #undef COMMANDS
@@ -119,7 +126,6 @@ static __printf_1_2 void dss_msg(const char* fmt,...)
 	va_end(argp);
 }
 
-/* TODO: Also consider number of inodes. */
 static int disk_space_low(void)
 {
 	struct disk_space ds;
@@ -144,7 +150,7 @@ static void dss_get_snapshot_list(struct snapshot_list *sl)
 	get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
 }
 
-static void compute_next_snapshot_time(void)
+static int next_snapshot_is_due(void)
 {
 	struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
 		tmp, diff;
@@ -155,7 +161,10 @@ static void compute_next_snapshot_time(void)
 	struct snapshot *s = NULL;
 	struct snapshot_list sl;
 
-	assert(snapshot_creation_status == SCS_READY);
+	gettimeofday(&now, NULL);
+	if (tv_diff(&next_snapshot_time, &now, NULL) > 0)
+		return 0;
+	assert(snapshot_creation_status == HS_READY);
 	current_snapshot_creation_time = 0;
 	dss_get_snapshot_list(&sl);
 	FOR_EACH_SNAPSHOT(s, i, &sl) {
@@ -171,50 +180,149 @@ static void compute_next_snapshot_time(void)
 	tmp.tv_sec = x;
 	tmp.tv_usec = 0;
 	ret = tv_diff(&unit_interval, &tmp, &diff); /* total sleep time per unit interval */
-	gettimeofday(&now, NULL);
-	if (ret < 0 || !s)
-		goto min_sleep;
-	tv_divide(wanted, &diff, &tmp); /* sleep time betweeen two snapshots */
-	diff.tv_sec = s->completion_time;
+	if (ret < 0 || !s) { /* unit_interval < tmp or no snapshot */
+		ret = 1;
+		goto out;
+	}
+
+	tv_divide(wanted, &diff, &tmp); /* sleep time between two snapshots */
+	diff.tv_sec = s->completion_time; /* completion time of the latest snapshot */
 	diff.tv_usec = 0;
 	tv_add(&diff, &tmp, &next_snapshot_time);
-	if (tv_diff(&now, &next_snapshot_time, NULL) < 0)
-		goto out;
-min_sleep:
-	next_snapshot_time = now;
-	next_snapshot_time.tv_sec += 60;
+	ret = (tv_diff(&now, &next_snapshot_time, &diff) < 0)? 0 : 1;
 out:
 	free_snapshot_list(&sl);
+	if (ret == 0)
+		DSS_DEBUG_LOG("next snapshot due in %lu seconds\n",
+			tv2ms(&diff) / 1000);
+	else
+		DSS_DEBUG_LOG("next snapshot: now\n");
+	return ret;
+}
+
+static int pre_create_hook(void)
+{
+	int ret, fds[3] = {0, 0, 0};
+
+	assert(snapshot_creation_status == HS_READY);
+	if (!conf.pre_create_hook_given) {
+		snapshot_creation_status = HS_PRE_SUCCESS;
+		return 0;
+	}
+	DSS_DEBUG_LOG("executing %s\n", conf.pre_create_hook_arg);
+	ret = dss_exec_cmdline_pid(&create_pid,
+		conf.pre_create_hook_arg, fds);
+	if (ret < 0)
+		return ret;
+	snapshot_creation_status = HS_PRE_RUNNING;
+	return ret;
 }
 
+static int pre_remove_hook(struct snapshot *s, const char *why)
+{
+	int ret, fds[3] = {0, 0, 0};
+	char *cmd;
+
+	if (!s)
+		return 0;
+	DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
+	assert(snapshot_removal_status == HS_READY);
+	assert(remove_pid == 0);
+	assert(!snapshot_currently_being_removed);
+
+	snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
+	*snapshot_currently_being_removed = *s;
+	snapshot_currently_being_removed->name = dss_strdup(s->name);
 
-static int remove_snapshot(struct snapshot *s)
+	if (!conf.pre_remove_hook_given) {
+		snapshot_removal_status = HS_PRE_SUCCESS;
+		return 0;
+	}
+	cmd = make_message("%s %s/%s", conf.pre_remove_hook_arg,
+		conf.dest_dir_arg, s->name);
+	DSS_DEBUG_LOG("executing %s\n", cmd);
+	ret = dss_exec_cmdline_pid(&remove_pid,
+		conf.pre_remove_hook_arg, fds);
+	free(cmd);
+	if (ret < 0)
+		return ret;
+	snapshot_removal_status = HS_PRE_RUNNING;
+	return ret;
+}
+
+static int exec_rm(void)
 {
+	struct snapshot *s = snapshot_currently_being_removed;
 	int fds[3] = {0, 0, 0};
-	assert(!rm_pid);
 	char *new_name = being_deleted_name(s);
-	int ret = dss_rename(s->name, new_name);
 	char *argv[] = {"rm", "-rf", new_name, NULL};
+	int ret;
 
+	assert(snapshot_removal_status == HS_PRE_SUCCESS);
+	assert(remove_pid == 0);
+
+	DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
+	ret = dss_rename(s->name, new_name);
 	if (ret < 0)
 		goto out;
-	DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
-	ret = dss_exec(&rm_pid, argv[0], argv, fds);
+	ret = dss_exec(&remove_pid, argv[0], argv, fds);
+	if (ret < 0)
+		goto out;
+	snapshot_removal_status = HS_RUNNING;
 out:
 	free(new_name);
 	return ret;
 }
 
+static int snapshot_is_being_created(struct snapshot *s)
+{
+	return s->creation_time == current_snapshot_creation_time;
+}
+
+static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
+{
+	struct snapshot *s;
+	int i;
+
+	DSS_DEBUG_LOG("looking for orphaned snapshots\n");
+	FOR_EACH_SNAPSHOT(s, i, sl) {
+		if (snapshot_is_being_created(s))
+			continue;
+		/*
+		 * We know that no rm is currently running, so if s is marked
+		 * as being deleted, a previously started rm must have failed.
+		 */
+		if (s->flags & SS_BEING_DELETED)
+			return s;
+
+		if (s->flags & SS_COMPLETE) /* good snapshot */
+			continue;
+		/*
+		 * This snapshot is incomplete and it is not the snapshot
+		 * currently being created. However, we must not remove it if
+		 * rsync is about to be restarted. As only the newest snapshot
+		 * can be restarted, this snapshot is orphaned if it is not the
+		 * newest snapshot or if we are not about to restart rsync.
+		 */
+		if (get_newest_snapshot(sl) != s)
+			return s;
+		if (snapshot_creation_status != HS_NEEDS_RESTART)
+			return s;
+	}
+	/* no orphaned snapshots */
+	return NULL;
+}
+
 /*
  * return: 0: no redundant snapshots, 1: rm process started, negative: error
  */
-static int remove_redundant_snapshot(struct snapshot_list *sl)
+static struct snapshot *find_redundant_snapshot(struct snapshot_list *sl)
 {
-	int ret, i, interval;
+	int i, interval;
 	struct snapshot *s;
 	unsigned missing = 0;
 
-	DSS_INFO_LOG("looking for intervals containing too many snapshots\n");
+	DSS_DEBUG_LOG("looking for intervals containing too many snapshots\n");
 	for (interval = conf.num_intervals_arg - 1; interval >= 0; interval--) {
 		unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 		unsigned num = sl->interval_count[interval];
@@ -231,6 +339,8 @@ static int remove_redundant_snapshot(struct snapshot_list *sl)
 		FOR_EACH_SNAPSHOT(s, i, sl) {
 			int64_t this_score;
 
+			if (snapshot_is_being_created(s))
+				continue;
 			//DSS_DEBUG_LOG("checking %s\n", s->name);
 			if (s->interval > interval) {
 				prev = s;
@@ -255,50 +365,38 @@ static int remove_redundant_snapshot(struct snapshot_list *sl)
 			prev = s;
 		}
 		assert(victim);
-		if (conf.dry_run_given) {
-			dss_msg("%s would be removed (interval = %i)\n",
-				victim->name, victim->interval);
-			continue;
-		}
-		ret = remove_snapshot(victim);
-		return ret < 0? ret : 1;
+		return victim;
 	}
-	return 0;
+	return NULL;
 }
 
-static int remove_outdated_snapshot(struct snapshot_list *sl)
+static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 {
-	int i, ret;
+	int i;
 	struct snapshot *s;
 
-	DSS_INFO_LOG("looking for snapshots belonging to intervals greater than %d\n",
+	DSS_DEBUG_LOG("looking for snapshots belonging to intervals greater than %d\n",
 		conf.num_intervals_arg);
 	FOR_EACH_SNAPSHOT(s, i, sl) {
-		if (s->interval <= conf.num_intervals_arg)
+		if (snapshot_is_being_created(s))
 			continue;
-		if (conf.dry_run_given) {
-			dss_msg("%s would be removed (interval = %i)\n",
-				s->name, s->interval);
+		if (s->interval <= conf.num_intervals_arg)
 			continue;
-		}
-		ret = remove_snapshot(s);
-		if (ret < 0)
-			return ret;
-		return 1;
+		return s;
 	}
-	return 0;
+	return NULL;
 }
 
-static int remove_oldest_snapshot(struct snapshot_list *sl)
+struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 {
 	struct snapshot *s = get_oldest_snapshot(sl);
 
 	if (!s) /* no snapshot found */
-		return 0;
+		return NULL;
 	DSS_INFO_LOG("oldest snapshot: %s\n", s->name);
-	if (s->creation_time == current_snapshot_creation_time)
-		return 0; /* do not remove the snapshot currently being created */
-	return remove_snapshot(s);
+	if (snapshot_is_being_created(s))
+		return NULL;
+	return s;
 }
 
 static int rename_incomplete_snapshot(int64_t start)
@@ -324,64 +422,86 @@ static int try_to_free_disk_space(int low_disk_space)
 {
 	int ret;
 	struct snapshot_list sl;
+	struct snapshot *victim;
+	struct timeval now;
+	const char *why;
 
+	gettimeofday(&now, NULL);
+	if (tv_diff(&next_removal_check, &now, NULL) > 0)
+		return 0;
+	if (!low_disk_space && conf.keep_redundant_given)
+		return 0;
 	dss_get_snapshot_list(&sl);
-	ret = remove_outdated_snapshot(&sl);
-	if (ret) /* error, or we are removing something */
-		goto out;
-	/* no outdated snapshot */
-	ret = remove_redundant_snapshot(&sl);
-	if (ret)
-		goto out;
+	why = "outdated";
+	victim = find_outdated_snapshot(&sl);
+	if (victim)
+		goto remove;
+	why = "redundant";
+	victim = find_redundant_snapshot(&sl);
+	if (victim)
+		goto remove;
+	/* try harder only if disk space is low */
 	ret = 0;
 	if (!low_disk_space)
 		goto out;
+	why = "orphaned";
+	victim = find_orphaned_snapshot(&sl);
+	if (victim)
+		goto remove;
 	DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
-	ret = remove_oldest_snapshot(&sl);
-	if (ret)
-		goto out;
+	victim = find_oldest_removable_snapshot(&sl);
+	if (victim)
+		goto remove;
 	DSS_CRIT_LOG("uhuhu: not enough disk space for a single snapshot\n");
-	ret= -ENOSPC;
+	ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
+	goto out;
+remove:
+	ret = pre_remove_hook(victim, why);
 out:
 	free_snapshot_list(&sl);
 	return ret;
 }
 
-static int pre_create_hook(void)
+static int post_create_hook(void)
 {
 	int ret, fds[3] = {0, 0, 0};
+	char *cmd;
 
-	if (!conf.pre_create_hook_given) {
-		snapshot_creation_status = SCS_PRE_HOOK_SUCCESS;
+	if (!conf.post_create_hook_given) {
+		snapshot_creation_status = HS_READY;
 		return 0;
 	}
-	DSS_NOTICE_LOG("executing %s\n", conf.pre_create_hook_arg);
-	ret = dss_exec_cmdline_pid(&pre_create_hook_pid,
-		conf.pre_create_hook_arg, fds);
+	cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
+		conf.dest_dir_arg, path_to_last_complete_snapshot);
+	DSS_NOTICE_LOG("executing %s\n", cmd);
+	ret = dss_exec_cmdline_pid(&create_pid, cmd, fds);
+	free(cmd);
 	if (ret < 0)
 		return ret;
-	snapshot_creation_status = SCS_PRE_HOOK_RUNNING;
+	snapshot_creation_status = HS_POST_RUNNING;
 	return ret;
 }
 
-static int post_create_hook(void)
+static int post_remove_hook(void)
 {
 	int ret, fds[3] = {0, 0, 0};
 	char *cmd;
+	struct snapshot *s = snapshot_currently_being_removed;
 
-	if (!conf.post_create_hook_given) {
-		snapshot_creation_status = SCS_READY;
-		compute_next_snapshot_time();
+	assert(s);
+
+	if (!conf.post_remove_hook_given) {
+		snapshot_removal_status = HS_READY;
 		return 0;
 	}
-	cmd = make_message("%s %s", conf.post_create_hook_arg,
-		path_to_last_complete_snapshot);
+	cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
+		conf.dest_dir_arg, s->name);
 	DSS_NOTICE_LOG("executing %s\n", cmd);
-	ret = dss_exec_cmdline_pid(&post_create_hook_pid, cmd, fds);
+	ret = dss_exec_cmdline_pid(&remove_pid, cmd, fds);
 	free(cmd);
 	if (ret < 0)
 		return ret;
-	snapshot_creation_status = SCS_POST_HOOK_RUNNING;
+	snapshot_removal_status = HS_POST_RUNNING;
 	return ret;
 }
 
@@ -393,23 +513,22 @@ static void kill_process(pid_t pid)
 	kill(pid, SIGTERM);
 }
 
-static void stop_rsync_process(void)
+static void stop_create_process(void)
 {
-	if (!rsync_pid || rsync_stopped)
+	if (!create_pid || create_process_stopped)
 		return;
-	kill(SIGSTOP, rsync_pid);
-	rsync_stopped = 1;
+	kill(SIGSTOP, create_pid);
+	create_process_stopped = 1;
 }
 
-static void restart_rsync_process(void)
+static void restart_create_process(void)
 {
-	if (!rsync_pid || !rsync_stopped)
+	if (!create_pid || !create_process_stopped)
 		return;
-	kill (SIGCONT, rsync_pid);
-	rsync_stopped = 0;
+	kill (SIGCONT, create_pid);
+	create_process_stopped = 0;
 }
 
-
 /**
  * Print a log message about the exit status of a child.
  */
@@ -461,23 +580,83 @@ static int wait_for_process(pid_t pid, int *status)
 	return ret;
 }
 
+static void handle_pre_remove_exit(int status)
+{
+	if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
+		snapshot_removal_status = HS_READY;
+		gettimeofday(&next_removal_check, NULL);
+		next_removal_check.tv_sec += 60;
+		return;
+	}
+	snapshot_removal_status = HS_PRE_SUCCESS;
+}
+
 static int handle_rm_exit(int status)
 {
-	rm_pid = 0;
-	if (!WIFEXITED(status))
+	if (!WIFEXITED(status)) {
+		snapshot_removal_status = HS_READY;
 		return -E_INVOLUNTARY_EXIT;
-	if (WEXITSTATUS(status))
+	}
+	if (WEXITSTATUS(status)) {
+		snapshot_removal_status = HS_READY;
 		return -E_BAD_EXIT_CODE;
+	}
+	if (conf.post_remove_hook_given)
+		snapshot_removal_status = HS_SUCCESS;
+	else
+		snapshot_removal_status = HS_READY;
 	return 1;
 }
 
-static int wait_for_rm_process(void)
+static void handle_post_remove_exit(void)
 {
-	int status, ret = wait_for_process(rm_pid, &status);
+	snapshot_removal_status = HS_READY;
+}
+
+static int handle_remove_exit(int status)
+{
+	int ret;
+	struct snapshot *s = snapshot_currently_being_removed;
 
+	assert(s);
+	switch (snapshot_removal_status) {
+	case HS_PRE_RUNNING:
+		handle_pre_remove_exit(status);
+		ret = 1;
+		break;
+	case HS_RUNNING:
+		ret = handle_rm_exit(status);
+		break;
+	case HS_POST_RUNNING:
+		handle_post_remove_exit();
+		ret = 1;
+		break;
+	default:
+		ret = -E_BUG;
+	}
+	if (snapshot_removal_status == HS_READY) {
+		free(s->name);
+		free(s);
+		snapshot_currently_being_removed = NULL;
+	}
+	remove_pid = 0;
+	return ret;
+}
+
+static int wait_for_remove_process(void)
+{
+	int status, ret;
+
+	assert(remove_pid);
+	assert(
+		snapshot_removal_status == HS_PRE_RUNNING ||
+		snapshot_removal_status == HS_RUNNING ||
+		snapshot_removal_status == HS_POST_RUNNING
+	);
+	ret = wait_for_process(remove_pid, &status);
 	if (ret < 0)
 		return ret;
-	return handle_rm_exit(status);
+	return handle_remove_exit(status);
 }
 
 static int handle_rsync_exit(int status)
@@ -485,51 +664,69 @@ static int handle_rsync_exit(int status)
 	int es, ret;
 
 	if (!WIFEXITED(status)) {
-		DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)rsync_pid);
+		DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
 		ret = -E_INVOLUNTARY_EXIT;
-		snapshot_creation_status = SCS_READY;
-		compute_next_snapshot_time();
+		snapshot_creation_status = HS_READY;
 		goto out;
 	}
 	es = WEXITSTATUS(status);
+	/*
+	 * Restart rsync on non-fatal errors:
+	 * 12: Error in rsync protocol data stream
+	 * 13: Errors with program diagnostics
+	 */
+	if (es == 12 || es == 13) {
+		DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
+			(int)create_pid, es);
+		snapshot_creation_status = HS_NEEDS_RESTART;
+		gettimeofday(&next_snapshot_time, NULL);
+		next_snapshot_time.tv_sec += 60;
+		ret = 1;
+		goto out;
+	}
 	if (es != 0 && es != 23 && es != 24) {
-		DSS_ERROR_LOG("rsync process %d returned %d\n", (int)rsync_pid, es);
+		DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
 		ret = -E_BAD_EXIT_CODE;
-		snapshot_creation_status = SCS_READY;
-		compute_next_snapshot_time();
+		snapshot_creation_status = HS_READY;
 		goto out;
 	}
 	ret = rename_incomplete_snapshot(current_snapshot_creation_time);
 	if (ret < 0)
 		goto out;
-	snapshot_creation_status = SCS_RSYNC_SUCCESS;
+	snapshot_creation_status = HS_SUCCESS;
 out:
-	rsync_pid = 0;
-	rsync_stopped = 0;
+	create_pid = 0;
+	create_process_stopped = 0;
 	return ret;
 }
 
 static int handle_pre_create_hook_exit(int status)
 {
 	int es, ret;
+	static int warn_count;
 
 	if (!WIFEXITED(status)) {
-		snapshot_creation_status = SCS_READY;
-		compute_next_snapshot_time();
+		snapshot_creation_status = HS_READY;
 		ret = -E_INVOLUNTARY_EXIT;
 		goto out;
 	}
 	es = WEXITSTATUS(status);
 	if (es) {
-		snapshot_creation_status = SCS_READY;
-		compute_next_snapshot_time();
-		ret = -E_BAD_EXIT_CODE;
+		if (!warn_count--) {
+			DSS_NOTICE_LOG("pre_create_hook %s returned %d\n",
+				conf.pre_create_hook_arg, es);
+			DSS_NOTICE_LOG("deferring snapshot creation...\n");
+			warn_count = 60; /* warn only once per hour */
+		}
+		snapshot_creation_status = HS_READY;
+		ret = 0;
 		goto out;
 	}
-	snapshot_creation_status = SCS_PRE_HOOK_SUCCESS;
+	warn_count = 0;
+	snapshot_creation_status = HS_PRE_SUCCESS;
 	ret = 1;
 out:
-	pre_create_hook_pid = 0;
+	create_pid = 0;
 	return ret;
 }
 
@@ -540,19 +737,30 @@ static int handle_sigchld(void)
 
 	if (ret <= 0)
 		return ret;
-	if (pid == rsync_pid)
-		return handle_rsync_exit(status);
-	if (pid == rm_pid)
-		return handle_rm_exit(status);
-	if (pid == pre_create_hook_pid)
-		return handle_pre_create_hook_exit(status);
-	if (pid == post_create_hook_pid) {
-		snapshot_creation_status = SCS_READY;
-		compute_next_snapshot_time();
-		return 1;
+
+	if (pid == create_pid) {
+		switch (snapshot_creation_status) {
+		case HS_PRE_RUNNING:
+			return handle_pre_create_hook_exit(status);
+		case HS_RUNNING:
+			return handle_rsync_exit(status);
+		case HS_POST_RUNNING:
+			snapshot_creation_status = HS_READY;
+			return 1;
+		default:
+			DSS_EMERG_LOG("BUG: create can't die in status %d\n",
+				snapshot_creation_status);
+			return -E_BUG;
+		}
+	}
+	if (pid == remove_pid) {
+		ret = handle_remove_exit(status);
+		if (ret < 0)
+			return ret;
+		return ret;
 	}
 	DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
-	exit(EXIT_FAILURE);
+	return -E_BUG;
 }
 
 static int check_config(void)
@@ -570,8 +778,7 @@ static int check_config(void)
 	return 1;
 }
 
-/* exits on errors */
-static void parse_config_file(int override)
+static int parse_config_file(int override)
 {
 	int ret;
 	char *config_file;
@@ -606,6 +813,15 @@ static void parse_config_file(int override)
 			.check_ambiguity = 0,
 			.print_errors = 1
 		};
+		if (override) { /* invalidate all rsync options */
+			int i;
+
+			for (i = 0; i < conf.rsync_option_given; i++) {
+				free(conf.rsync_option_arg[i]);
+				conf.rsync_option_arg[i] = NULL;
+			}
+			conf.rsync_option_given = 0;
+		}
 		cmdline_parser_config_file(config_file, &conf, &params);
 	}
 	ret = check_config();
@@ -627,36 +843,33 @@ static void parse_config_file(int override)
 		logfile = open_log(conf.logfile_arg);
 		log_welcome(conf.loglevel_arg);
 	}
-	DSS_EMERG_LOG("loglevel: %d\n", conf.loglevel_arg);
+	DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 //	cmdline_parser_dump(logfile? logfile : stdout, &conf);
 out:
 	free(config_file);
-	if (ret >= 0)
-		return;
-	DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
-	exit(EXIT_FAILURE);
+	if (ret < 0)
+		DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
+	return ret;
 }
 
-static void change_to_dest_dir(void)
+static int change_to_dest_dir(void)
 {
-	int ret;
-
 	DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
-	ret = dss_chdir(conf.dest_dir_arg);
-	if (ret >= 0)
-		return;
-	DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
-	exit(EXIT_FAILURE);
+	return dss_chdir(conf.dest_dir_arg);
 }
 
-static void handle_sighup(void)
+static int handle_sighup(void)
 {
+	int ret;
+
 	DSS_NOTICE_LOG("SIGHUP\n");
-	parse_config_file(1);
-	change_to_dest_dir();
+	ret = parse_config_file(1);
+	if (ret < 0)
+		return ret;
+	return change_to_dest_dir();
 }
 
-static void handle_signal(void)
+static int handle_signal(void)
 {
 	int sig, ret = next_signal();
 
@@ -666,13 +879,13 @@ static void handle_signal(void)
 	switch (sig) {
 	case SIGINT:
 	case SIGTERM:
-		restart_rsync_process();
-		kill_process(rsync_pid);
-		kill_process(rm_pid);
-		exit(EXIT_FAILURE);
+		restart_create_process();
+		kill_process(create_pid);
+		kill_process(remove_pid);
+		ret = -E_SIGNAL;
+		break;
 	case SIGHUP:
-		handle_sighup();
-		ret = 1;
+		ret = handle_sighup();
 		break;
 	case SIGCHLD:
 		ret = handle_sigchld();
@@ -681,6 +894,22 @@ static void handle_signal(void)
 out:
 	if (ret < 0)
 		DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
+	return ret;
+}
+
+/*
+ * We can not use rsync locally if the local user is different from the remote
+ * user or if the src dir is not on the local host (or both).
+ */
+static int use_rsync_locally(char *logname)
+{
+	char *h = conf.remote_host_arg;
+
+	if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
+		return 0;
+	if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
+		return 0;
+	return 1;
 }
 
 static void create_rsync_argv(char ***argv, int64_t *num)
@@ -706,7 +935,7 @@ static void create_rsync_argv(char ***argv, int64_t *num)
 	} else
 		DSS_INFO_LOG("no previous snapshot found\n");
 	logname = dss_logname();
-	if (conf.remote_user_given && !strcmp(conf.remote_user_arg, logname))
+	if (use_rsync_locally(logname))
 		(*argv)[i++] = dss_strdup(conf.source_dir_arg);
 	else
 		(*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
@@ -723,6 +952,9 @@ static void create_rsync_argv(char ***argv, int64_t *num)
 static void free_rsync_argv(char **argv)
 {
 	int i;
+
+	if (!argv)
+		return;
 	for (i = 0; argv[i]; i++)
 		free(argv[i]);
 	free(argv);
@@ -736,83 +968,113 @@ static int create_snapshot(char **argv)
 	name = incomplete_name(current_snapshot_creation_time);
 	DSS_NOTICE_LOG("creating new snapshot %s\n", name);
 	free(name);
-	ret = dss_exec(&rsync_pid, argv[0], argv, fds);
+	ret = dss_exec(&create_pid, argv[0], argv, fds);
 	if (ret < 0)
 		return ret;
-	snapshot_creation_status = SCS_RSYNC_RUNNING;
+	snapshot_creation_status = HS_RUNNING;
 	return ret;
 }
 
 static int select_loop(void)
 {
 	int ret;
-	struct timeval tv = {.tv_sec = 0, .tv_usec = 0};
+	/* check every 60 seconds for free disk space */
+	struct timeval tv;
+	char **rsync_argv = NULL;
 
 	for (;;) {
 		fd_set rfds;
 		int low_disk_space;
-		char **rsync_argv;
-		struct timeval now, *tvp = &tv;
-
-		if (rsync_pid)
-			tv.tv_sec = 60; /* check every 60 seconds for free disk space */
-		else if (rm_pid)
-			tvp = NULL; /* sleep until rm process dies */
+		struct timeval *tvp;
+
+		if (remove_pid)
+			tvp = NULL; /* sleep until rm hook/process dies */
+		else { /* sleep one minute */
+			tv.tv_sec = 60;
+			tv.tv_usec = 0;
+			tvp = &tv;
+		}
 		FD_ZERO(&rfds);
 		FD_SET(signal_pipe, &rfds);
-		DSS_DEBUG_LOG("tvp: %p, tv_sec: %lu\n", tvp, (long unsigned) tv.tv_sec);
+		DSS_DEBUG_LOG("tvp: %p, tv_sec : %lu\n", tvp, (long unsigned) tv.tv_sec);
 		ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
 		if (ret < 0)
-			return ret;
-		if (FD_ISSET(signal_pipe, &rfds))
-			handle_signal();
-		if (rm_pid)
+			goto out;
+		if (FD_ISSET(signal_pipe, &rfds)) {
+			ret = handle_signal();
+			if (ret < 0)
+				goto out;
+		}
+		if (remove_pid)
+			continue;
+		if (snapshot_removal_status == HS_PRE_SUCCESS) {
+			ret = exec_rm();
+			if (ret < 0)
+				goto out;
 			continue;
+		}
+		if (snapshot_removal_status == HS_SUCCESS) {
+			ret = post_remove_hook();
+			if (ret < 0)
+				goto out;
+			continue;
+		}
 		ret = disk_space_low();
 		if (ret < 0)
-			break;
+			goto out;
 		low_disk_space = ret;
 		ret = try_to_free_disk_space(low_disk_space);
 		if (ret < 0)
-			break;
-		if (rm_pid) {
-			stop_rsync_process();
+			goto out;
+		if (snapshot_removal_status != HS_READY) {
+			stop_create_process();
 			continue;
 		}
-		restart_rsync_process();
-		gettimeofday(&now, NULL);
-		if (tv_diff(&next_snapshot_time, &now, &tv) > 0)
-			continue;
+		restart_create_process();
 		switch (snapshot_creation_status) {
-		case SCS_READY:
+		case HS_READY:
+			if (!next_snapshot_is_due())
+				continue;
 			ret = pre_create_hook();
 			if (ret < 0)
 				goto out;
 			continue;
-		case SCS_PRE_HOOK_RUNNING:
+		case HS_PRE_RUNNING:
+		case HS_RUNNING:
+		case HS_POST_RUNNING:
 			continue;
-		case SCS_PRE_HOOK_SUCCESS:
+		case HS_PRE_SUCCESS:
+			free_rsync_argv(rsync_argv);
 			create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
+			/* fall through */
+		case HS_NEEDS_RESTART:
 			ret = create_snapshot(rsync_argv);
-			free_rsync_argv(rsync_argv);
 			if (ret < 0)
 				goto out;
 			continue;
-		case SCS_RSYNC_RUNNING:
-			continue;
-		case SCS_RSYNC_SUCCESS:
+		case HS_SUCCESS:
 			ret = post_create_hook();
 			if (ret < 0)
 				goto out;
 			continue;
-		case SCS_POST_HOOK_RUNNING:
-			continue;
 		}
 	}
 out:
 	return ret;
 }
 
+static void exit_hook(int exit_code)
+{
+	int fds[3] = {0, 0, 0};
+	char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
+	pid_t pid;
+
+	if (!conf.exit_hook_given)
+		return;
+	DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
+	dss_exec(&pid, conf.exit_hook_arg, argv, fds);
+}
+
 static int com_run(void)
 {
 	int ret;
@@ -824,46 +1086,72 @@ static int com_run(void)
 	ret = install_sighandler(SIGHUP);
 	if (ret < 0)
 		return ret;
-	compute_next_snapshot_time();
-	return select_loop();
+	ret = select_loop();
+	if (ret >= 0) /* impossible */
+		ret = -E_BUG;
+	exit_hook(ret);
+	return ret;
 }
 
 static int com_prune(void)
 {
 	int ret;
 	struct snapshot_list sl;
+	struct snapshot *victim;
 	struct disk_space ds;
+	const char *why;
 
 	ret = get_disk_space(".", &ds);
 	if (ret < 0)
 		return ret;
 	log_disk_space(&ds);
-	for (;;) {
-		dss_get_snapshot_list(&sl);
-		ret = remove_outdated_snapshot(&sl);
-		free_snapshot_list(&sl);
-		if (ret < 0)
-			return ret;
-		if (!ret)
-			break;
-		ret = wait_for_rm_process();
-		if (ret < 0)
-			goto out;
+	dss_get_snapshot_list(&sl);
+	why = "outdated";
+	victim = find_outdated_snapshot(&sl);
+	if (victim)
+		goto rm;
+	why = "redundant";
+	victim = find_redundant_snapshot(&sl);
+	if (victim)
+		goto rm;
+	ret = 0;
+	goto out;
+rm:
+	if (conf.dry_run_given) {
+		dss_msg("%s snapshot %s (interval = %i)\n",
+			why, victim->name, victim->interval);
+		ret = 0;
+		goto out;
 	}
-	for (;;) {
-		dss_get_snapshot_list(&sl);
-		ret = remove_redundant_snapshot(&sl);
-		free_snapshot_list(&sl);
-		if (ret < 0)
-			return ret;
-		if (!ret)
-			break;
-		ret = wait_for_rm_process();
+	ret = pre_remove_hook(victim, why);
+	if (ret < 0)
+		goto out;
+	if (snapshot_removal_status == HS_PRE_RUNNING) {
+		ret = wait_for_remove_process();
 		if (ret < 0)
 			goto out;
+		if (snapshot_removal_status != HS_PRE_SUCCESS)
+			goto out;
 	}
-	return 1;
+	ret = exec_rm();
+	if (ret < 0)
+		goto out;
+	ret = wait_for_remove_process();
+	if (ret < 0)
+		goto out;
+	if (snapshot_removal_status != HS_SUCCESS)
+		goto out;
+	ret = post_remove_hook();
+	if (ret < 0)
+		goto out;
+	if (snapshot_removal_status != HS_POST_RUNNING)
+		goto out;
+	ret = wait_for_remove_process();
+	if (ret < 0)
+		goto out;
+	ret = 1;
 out:
+	free_snapshot_list(&sl);
 	return ret;
 }
 
@@ -890,27 +1178,27 @@ static int com_create(void)
 	ret = pre_create_hook();
 	if (ret < 0)
 		return ret;
-	if (pre_create_hook_pid) {
-		ret = wait_for_process(pre_create_hook_pid, &status);
+	if (create_pid) {
+		ret = wait_for_process(create_pid, &status);
 		if (ret < 0)
 			return ret;
 		ret = handle_pre_create_hook_exit(status);
-		if (ret < 0)
+		if (ret <= 0) /* error, or pre-create failed */
 			return ret;
 	}
 	create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
 	ret = create_snapshot(rsync_argv);
 	if (ret < 0)
 		goto out;
-	ret = wait_for_process(rsync_pid, &status);
+	ret = wait_for_process(create_pid, &status);
 	if (ret < 0)
 		goto out;
 	ret = handle_rsync_exit(status);
 	if (ret < 0)
 		goto out;
 	post_create_hook();
-	if (post_create_hook_pid)
-		ret = wait_for_process(post_create_hook_pid, &status);
+	if (create_pid)
+		ret = wait_for_process(create_pid, &status);
 out:
 	free_rsync_argv(rsync_argv);
 	return ret;
@@ -923,13 +1211,17 @@ static int com_ls(void)
 	struct snapshot *s;
 
 	dss_get_snapshot_list(&sl);
-	FOR_EACH_SNAPSHOT(s, i, &sl)
-		dss_msg("%u\t%s\n", s->interval, s->name);
+	FOR_EACH_SNAPSHOT(s, i, &sl) {
+		int64_t d = 0;
+		if (s->flags & SS_COMPLETE)
+			d = (s->completion_time - s->creation_time) / 60;
+		dss_msg("%u\t%s\t%3" PRId64 ":%02" PRId64 "\n", s->interval, s->name, d/60, d%60);
+	};
 	free_snapshot_list(&sl);
 	return 1;
 }
 
-static void setup_signal_handling(void)
+static int setup_signal_handling(void)
 {
 	int ret;
 
@@ -937,17 +1229,11 @@ static void setup_signal_handling(void)
 	signal_pipe = signal_init(); /* always successful */
 	ret = install_sighandler(SIGINT);
 	if (ret < 0)
-		goto err;
+		return ret;
 	ret = install_sighandler(SIGTERM);
 	if (ret < 0)
-		goto err;
-	ret = install_sighandler(SIGCHLD);
-	if (ret < 0)
-		goto err;
-	return;
-err:
-	DSS_EMERG_LOG("could not install signal handlers\n");
-	exit(EXIT_FAILURE);
+		return ret;
+	return install_sighandler(SIGCHLD);
 }
 
 /**
@@ -968,13 +1254,19 @@ int main(int argc, char **argv)
 	};
 
 	cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
-	parse_config_file(0);
-
+	ret = parse_config_file(0);
+	if (ret < 0)
+		goto out;
 	if (conf.daemon_given)
 		daemon_init();
-	change_to_dest_dir();
-	setup_signal_handling();
+	ret = change_to_dest_dir();
+	if (ret < 0)
+		goto out;
+	ret = setup_signal_handling();
+	if (ret < 0)
+		goto out;
 	ret = call_command_handler();
+out:
 	if (ret < 0)
 		DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 	exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);