X-Git-Url: http://git.tuebingen.mpg.de/?p=dss.git;a=blobdiff_plain;f=dss.c;h=40eccab2b993ccab5c2f83f85fe6e06c664c40c1;hp=0b6e77e3fd8850dd76f2d87dfb2a74abe67629a8;hb=04bcb8e95d4afd843bcbb98cb01eb6c19a990446;hpb=bd0d0f91cfa537c4a74fc781811589bcef720436

diff --git a/dss.c b/dss.c
index 0b6e77e..40eccab 100644
--- a/dss.c
+++ b/dss.c
@@ -1,3 +1,8 @@
+/*
+ * Copyright (C) 2008-2009 Andre Noll <maan@systemlinux.org>
+ *
+ * Licensed under the GPL v2. For licencing details see COPYING.
+ */
 #include <string.h>
 #include <stdlib.h>
 #include <stdarg.h>
@@ -27,47 +32,35 @@
 #include "signal.h"
 #include "df.h"
 #include "time.h"
+#include "snap.h"
 
-
-struct gengetopt_args_info conf;
-char *dss_error_txt = NULL;
+/** Command line and config file options. */
+static struct gengetopt_args_info conf;
+/** Non-NULL if we log to a file. */
 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. */
-struct timeval next_snapshot_time;
-/** The pid of the pre-create hook. */
-pid_t pre_create_hook_pid;
-/** The pid of the post-create hook. */
-pid_t post_create_hook_pid;
-
-/* Creation time of the snapshot currently being created. */
-int64_t current_snapshot_creation_time;
-
+static struct timeval next_snapshot_time;
+/** 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;
-
-enum {
-	/** We are ready to take the next snapshot. */
-	SCS_READY,
-	/** The pre-creation hook has been started. */
-	SCS_PRE_HOOK_RUNNING,
-	/** The pre-creation hook exited successfully. */
-	SCS_PRE_HOOK_SUCCESS,
-	/** The rsync process is running. */
-	SCS_RSYNC_RUNNING,
-	/** The rsync process exited successfully. */
-	SCS_RSYNC_SUCCESS,
-	/** The post-create hook has been started- */
-	SCS_POST_HOOK_RUNNING,
-};
-
-static unsigned snapshot_creation_status;
+static char *name_of_reference_snapshot;
+/** \sa \ref snap.h for details. */
+enum hook_status snapshot_creation_status;
+/** \sa \ref snap.h for details. */
+enum hook_status snapshot_removal_status;
 
 
 DEFINE_DSS_ERRLIST;
@@ -79,40 +72,27 @@ DEFINE_DSS_ERRLIST;
 	COMMAND(create) \
 	COMMAND(prune) \
 	COMMAND(run)
-#define COMMAND(x) int com_ ##x(void);
+#define COMMAND(x) static int com_ ##x(void);
 COMMANDS
 #undef COMMAND
 #define COMMAND(x) if (conf.x ##_given) return com_ ##x();
-int call_command_handler(void)
+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
 
-/*
- * complete, not being deleted: 1204565370-1204565371.Sun_Mar_02_2008_14_33-Sun_Mar_02_2008_14_43
- * complete, being deleted: 1204565370-1204565371.being_deleted
- * incomplete, not being deleted: 1204565370-incomplete
- * incomplete, being deleted: 1204565370-incomplete.being_deleted
+/**
+ * The log function of dss.
+ *
+ * \param ll Loglevel.
+ * \param fml Usual format string.
+ *
+ * All DSS_XXX_LOG() macros use this function.
  */
-enum snapshot_status_flags {
-	/** The rsync process terminated successfully. */
-	SS_COMPLETE = 1,
-	/** The rm process is running to remove this snapshot. */
-	SS_BEING_DELETED = 2,
-};
-
-struct snapshot {
-	char *name;
-	int64_t creation_time;
-	int64_t completion_time;
-	enum snapshot_status_flags flags;
-	unsigned interval;
-};
-
 __printf_2_3 void dss_log(int ll, const char* fmt,...)
 {
 	va_list argp;
@@ -138,7 +118,7 @@ __printf_2_3 void dss_log(int ll, const char* fmt,...)
 /**
  * Print a message either to stdout or to the log file.
  */
-__printf_1_2 void dss_msg(const char* fmt,...)
+static __printf_1_2 void dss_msg(const char* fmt,...)
 {
 	FILE *outfd = conf.daemon_given? logfile : stdout;
 	va_list argp;
@@ -147,313 +127,212 @@ __printf_1_2 void dss_msg(const char* fmt,...)
 	va_end(argp);
 }
 
-/**
- * Return the desired number of snapshots of an interval.
- */
-unsigned num_snapshots(int interval)
+static int disk_space_low(void)
 {
-	unsigned n;
+	struct disk_space ds;
+	int ret = get_disk_space(".", &ds);
 
-	assert(interval >= 0);
+	if (ret < 0)
+		return ret;
+	if (conf.min_free_mb_arg)
+		if (ds.free_mb < conf.min_free_mb_arg)
+			return 1;
+	if (conf.min_free_percent_arg)
+		if (ds.percent_free < conf.min_free_percent_arg)
+			return 1;
+	if (conf.min_free_percent_inodes_arg)
+		if (ds.percent_free_inodes < conf.min_free_percent_inodes_arg)
+			return 1;
+	return 0;
+}
 
-	if (interval >= conf.num_intervals_arg)
-		return 0;
-	n = conf.num_intervals_arg - interval - 1;
-	return 1 << n;
+static void dss_get_snapshot_list(struct snapshot_list *sl)
+{
+	get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
 }
 
-/* return: Whether dirname is a snapshot directory (0: no, 1: yes) */
-int is_snapshot(const char *dirname, int64_t now, struct snapshot *s)
+static int next_snapshot_is_due(void)
 {
+	struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
+		tmp, diff;
+	int64_t x = 0;
+	unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
+		num_complete_snapshots = 0;
 	int i, ret;
-	char *dash, *dot, *tmp;
-	int64_t num;
+	struct snapshot *s = NULL;
+	struct snapshot_list sl;
 
-	assert(dirname);
-	dash = strchr(dirname, '-');
-	if (!dash || !dash[1] || dash == dirname)
-		return 0;
-	for (i = 0; dirname[i] != '-'; i++)
-		if (!isdigit(dirname[i]))
-			return 0;
-	tmp = dss_strdup(dirname);
-	tmp[i] = '\0';
-	ret = dss_atoi64(tmp, &num);
-	free(tmp);
-	if (ret < 0) {
-		free(dss_error_txt);
-		return 0;
-	}
-	assert(num >= 0);
-	if (num > now)
+	gettimeofday(&now, NULL);
+	if (tv_diff(&next_snapshot_time, &now, NULL) > 0)
 		return 0;
-	s->creation_time = num;
-	//DSS_DEBUG_LOG("%s start time: %lli\n", dirname, (long long)s->creation_time);
-	s->interval = (long long) ((now - s->creation_time)
-		/ conf.unit_interval_arg / 24 / 3600);
-	if (!strcmp(dash + 1, "incomplete")) {
-		s->completion_time = -1;
-		s->flags = 0; /* neither complete, nor being deleted */
-		goto success;
-	}
-	if (!strcmp(dash + 1, "incomplete.being_deleted")) {
-		s->completion_time = -1;
-		s->flags = SS_BEING_DELETED; /* mot cpmplete, being deleted */
-		goto success;
+	assert(snapshot_creation_status == HS_READY);
+	current_snapshot_creation_time = 0;
+	dss_get_snapshot_list(&sl);
+	FOR_EACH_SNAPSHOT(s, i, &sl) {
+		if (!(s->flags & SS_COMPLETE))
+			continue;
+		num_complete_snapshots++;
+		x += s->completion_time - s->creation_time;
 	}
-	tmp = dash + 1;
-	dot = strchr(tmp, '.');
-	if (!dot || !dot[1] || dot == tmp)
-		return 0;
-	for (i = 0; tmp[i] != '.'; i++)
-		if (!isdigit(tmp[i]))
-			return 0;
-	tmp = dss_strdup(dash + 1);
-	tmp[i] = '\0';
-	ret = dss_atoi64(tmp, &num);
-	free(tmp);
-	if (ret < 0) {
-		free(dss_error_txt);
-		return 0;
+	assert(x >= 0);
+	if (num_complete_snapshots)
+		x /= num_complete_snapshots; /* avg time to create one snapshot */
+	x *= wanted; /* time to create all snapshots in interval 0 */
+	tmp.tv_sec = x;
+	tmp.tv_usec = 0;
+	ret = tv_diff(&unit_interval, &tmp, &diff); /* total sleep time per unit interval */
+	if (ret < 0 || !s) { /* unit_interval < tmp or no snapshot */
+		ret = 1;
+		goto out;
 	}
-	if (num > now)
-		return 0;
-	s->completion_time = num;
-	s->flags = SS_COMPLETE;
-	if (!strcmp(dot + 1, "being_deleted"))
-		s->flags |= SS_BEING_DELETED;
-success:
-	s->name = dss_strdup(dirname);
-	return 1;
-}
 
-int64_t get_current_time(void)
-{
-	time_t now;
-	time(&now);
-	DSS_DEBUG_LOG("now: %lli\n", (long long) now);
-	return (int64_t)now;
+	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);
+	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;
 }
 
-char *incomplete_name(int64_t start)
+static int pre_create_hook(void)
 {
-	return make_message("%lli-incomplete", (long long)start);
-}
+	int ret, fds[3] = {0, 0, 0};
 
-char *being_deleted_name(struct snapshot *s)
-{
-	if (s->flags & SS_COMPLETE)
-		return make_message("%lli-%lli.being_deleted",
-			(long long)s->creation_time,
-			(long long)s->completion_time);
-	return make_message("%lli-incomplete.being_deleted",
-		(long long)s->creation_time);
+	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;
 }
 
-int complete_name(int64_t start, int64_t end, char **result)
+static int pre_remove_hook(struct snapshot *s, const char *why)
 {
-	struct tm start_tm, end_tm;
-	time_t *start_seconds = (time_t *) (uint64_t *)&start; /* STFU, gcc */
-	time_t *end_seconds = (time_t *) (uint64_t *)&end; /* STFU, gcc */
-	char start_str[200], end_str[200];
-
-	if (!localtime_r(start_seconds, &start_tm))
-		return -E_LOCALTIME;
-	if (!localtime_r(end_seconds, &end_tm))
-		return -E_LOCALTIME;
-	if (!strftime(start_str, sizeof(start_str), "%a_%b_%d_%Y_%H_%M_%S", &start_tm))
-		return -E_STRFTIME;
-	if (!strftime(end_str, sizeof(end_str), "%a_%b_%d_%Y_%H_%M_%S", &end_tm))
-		return -E_STRFTIME;
-	*result = make_message("%lli-%lli.%s-%s", (long long) start, (long long) end,
-		start_str, end_str);
-	return 1;
-}
-
-struct snapshot_list {
-	int64_t now;
-	unsigned num_snapshots;
-	unsigned array_size;
-	struct snapshot **snapshots;
-	/**
-	 * Array of size num_intervals + 1
-	 *
-	 * It contains the number of snapshots in each interval. interval_count[num_intervals]
-	 * is the number of snapshots which belong to any interval greater than num_intervals.
-	 */
-	unsigned *interval_count;
-};
+	int ret, fds[3] = {0, 0, 0};
+	char *cmd;
 
-#define FOR_EACH_SNAPSHOT(s, i, sl) \
-	for ((i) = 0; (i) < (sl)->num_snapshots && ((s) = (sl)->snapshots[(i)]); (i)++)
+	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);
 
-#define FOR_EACH_SNAPSHOT_REVERSE(s, i, sl) \
-	for ((i) = (sl)->num_snapshots; (i) > 0 && ((s) = (sl)->snapshots[(i - 1)]); (i)--)
+	snapshot_currently_being_removed = dss_malloc(sizeof(struct snapshot));
+	*snapshot_currently_being_removed = *s;
+	snapshot_currently_being_removed->name = dss_strdup(s->name);
 
-static inline struct snapshot *oldest_snapshot(struct snapshot_list *sl)
-{
-	if (!sl->num_snapshots)
-		return NULL;
-	return sl->snapshots[0];
+	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;
 }
 
-#define NUM_COMPARE(x, y) ((int)((x) < (y)) - (int)((x) > (y)))
-
-static int compare_snapshots(const void *a, const void *b)
+static int exec_rm(void)
 {
-	struct snapshot *s1 = *(struct snapshot **)a;
-	struct snapshot *s2 = *(struct snapshot **)b;
-	return NUM_COMPARE(s2->creation_time, s1->creation_time);
-}
-
-/** Compute the minimum of \a a and \a b. */
-#define DSS_MIN(a,b) ((a) < (b) ? (a) : (b))
+	struct snapshot *s = snapshot_currently_being_removed;
+	int fds[3] = {0, 0, 0};
+	char *new_name = being_deleted_name(s);
+	char *argv[] = {"rm", "-rf", new_name, NULL};
+	int ret;
 
-int add_snapshot(const char *dirname, void *private)
-{
-	struct snapshot_list *sl = private;
-	struct snapshot s;
-	int ret = is_snapshot(dirname, sl->now, &s);
+	assert(snapshot_removal_status == HS_PRE_SUCCESS);
+	assert(remove_pid == 0);
 
-	if (!ret)
-		return 1;
-	if (sl->num_snapshots >= sl->array_size) {
-		sl->array_size = 2 * sl->array_size + 1;
-		sl->snapshots = dss_realloc(sl->snapshots,
-			sl->array_size * sizeof(struct snapshot *));
-	}
-	sl->snapshots[sl->num_snapshots] = dss_malloc(sizeof(struct snapshot));
-	*(sl->snapshots[sl->num_snapshots]) = s;
-	sl->interval_count[DSS_MIN(s.interval, conf.num_intervals_arg)]++;
-	sl->num_snapshots++;
-	return 1;
+	DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
+	ret = dss_rename(s->name, new_name);
+	if (ret < 0)
+		goto out;
+	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;
 }
 
-void get_snapshot_list(struct snapshot_list *sl)
+static int snapshot_is_being_created(struct snapshot *s)
 {
-	sl->now = get_current_time();
-	sl->num_snapshots = 0;
-	sl->array_size = 0;
-	sl->snapshots = NULL;
-	sl->interval_count = dss_calloc((conf.num_intervals_arg + 1) * sizeof(unsigned));
-	for_each_subdir(add_snapshot, sl);
-	qsort(sl->snapshots, sl->num_snapshots, sizeof(struct snapshot *),
-		compare_snapshots);
+	return s->creation_time == current_snapshot_creation_time;
 }
 
-void free_snapshot_list(struct snapshot_list *sl)
+static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
 {
-	int i;
 	struct snapshot *s;
+	int i;
 
+	DSS_DEBUG_LOG("looking for orphaned snapshots\n");
 	FOR_EACH_SNAPSHOT(s, i, sl) {
-		free(s->name);
-		free(s);
-	}
-	free(sl->interval_count);
-	sl->interval_count = NULL;
-	free(sl->snapshots);
-	sl->snapshots = NULL;
-	sl->num_snapshots = 0;
-}
-
-void stop_rsync_process(void)
-{
-	if (!rsync_pid || rsync_stopped)
-		return;
-	kill(SIGSTOP, rsync_pid);
-	rsync_stopped = 1;
-}
-
-void restart_rsync_process(void)
-{
-	if (!rsync_pid || !rsync_stopped)
-		return;
-	kill (SIGCONT, rsync_pid);
-	rsync_stopped = 0;
-}
-
-/**
- * Print a log message about the exit status of a child.
- */
-void log_termination_msg(pid_t pid, int status)
-{
-	if (WIFEXITED(status))
-		DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
-			WEXITSTATUS(status));
-	else if (WIFSIGNALED(status))
-		DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
-			WTERMSIG(status));
-	else
-		DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
-}
-
-int wait_for_process(pid_t pid, int *status)
-{
-	int ret;
-
-	DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
-	for (;;) {
-		fd_set rfds;
-
-		FD_ZERO(&rfds);
-		FD_SET(signal_pipe, &rfds);
-		ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
-		if (ret < 0)
-			break;
-		ret = next_signal();
-		if (!ret)
+		if (snapshot_is_being_created(s))
 			continue;
-		if (ret == SIGCHLD) {
-			ret = waitpid(pid, status, 0);
-			if (ret >= 0)
-				break;
-			if (errno != EINTR) { /* error */
-				ret = -ERRNO_TO_DSS_ERROR(errno);
-				break;
-			}
-		}
-		/* SIGINT or SIGTERM */
-		DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
-		kill(pid, SIGTERM);
+		/*
+		 * 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;
 	}
-	if (ret < 0)
-		DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
-	else
-		log_termination_msg(pid, *status);
-	return ret;
+	/* no orphaned snapshots */
+	return NULL;
 }
 
-int remove_snapshot(struct snapshot *s)
+static int is_reference_snapshot(struct snapshot *s)
 {
-	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};
-
-	if (ret < 0)
-		goto out;
-	DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
-	stop_rsync_process();
-	ret = dss_exec(&rm_pid, argv[0], argv, fds);
-out:
-	free(new_name);
-	return ret;
+	if (!name_of_reference_snapshot)
+		return 0;
+	return strcmp(s->name, name_of_reference_snapshot)? 0 : 1;
 }
 
 /*
  * return: 0: no redundant snapshots, 1: rm process started, negative: error
  */
-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 = num_snapshots(interval);
+		unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
 		unsigned num = sl->interval_count[interval];
 		struct snapshot *victim = NULL, *prev = NULL;
 		int64_t score = LONG_MAX;
@@ -468,6 +347,10 @@ 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;
+			if (is_reference_snapshot(s))
+				continue;
 			//DSS_DEBUG_LOG("checking %s\n", s->name);
 			if (s->interval > interval) {
 				prev = s;
@@ -492,86 +375,433 @@ 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;
 }
 
-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 (is_reference_snapshot(s))
 			continue;
-		}
-		ret = remove_snapshot(s);
-		if (ret < 0)
-			return ret;
-		return 1;
+		if (s->interval <= conf.num_intervals_arg)
+			continue;
+		return s;
 	}
-	return 0;
+	return NULL;
 }
 
-int handle_rm_exit(int status)
+struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 {
-	rm_pid = 0;
-	if (!WIFEXITED(status))
-		return -E_INVOLUNTARY_EXIT;
-	if (WEXITSTATUS(status))
-		return -E_BAD_EXIT_CODE;
-	return 1;
+	int i;
+	struct snapshot *s;
+	FOR_EACH_SNAPSHOT(s, i, sl) {
+		if (snapshot_is_being_created(s))
+			continue;
+		if (is_reference_snapshot(s))
+			continue;
+		DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
+		return s;
+	}
+	return NULL;
 }
 
-int wait_for_rm_process(void)
+static int rename_incomplete_snapshot(int64_t start)
 {
-	int status, ret = wait_for_process(rm_pid, &status);
+	char *old_name;
+	int ret;
 
+	free(path_to_last_complete_snapshot);
+	ret = complete_name(start, get_current_time(),
+		&path_to_last_complete_snapshot);
 	if (ret < 0)
 		return ret;
-	return handle_rm_exit(status);
+	old_name = incomplete_name(start);
+	ret = dss_rename(old_name, path_to_last_complete_snapshot);
+	if (ret >= 0)
+		DSS_NOTICE_LOG("%s -> %s\n", old_name,
+			path_to_last_complete_snapshot);
+	free(old_name);
+	return ret;
 }
 
-void kill_process(pid_t pid)
+static int try_to_free_disk_space(int low_disk_space)
 {
-	if (!pid)
-		return;
-	DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
-	kill(pid, SIGTERM);
+	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 = 0;
+	if (!low_disk_space && sl.num_snapshots <= 1)
+		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 */
+	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");
+	victim = find_oldest_removable_snapshot(&sl);
+	if (victim)
+		goto remove;
+	DSS_CRIT_LOG("uhuhu: disk space low and nothing to remove\n");
+	ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
+	goto out;
+remove:
+	ret = pre_remove_hook(victim, why);
+out:
+	free_snapshot_list(&sl);
+	return ret;
 }
 
-int check_config(void)
+static int post_create_hook(void)
 {
-	if (conf.unit_interval_arg <= 0) {
-		DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
-		return -E_INVALID_NUMBER;
-	}
-	DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
-	if (conf.num_intervals_arg <= 0) {
-		DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
-		return -E_INVALID_NUMBER;
+	int ret, fds[3] = {0, 0, 0};
+	char *cmd;
+
+	if (!conf.post_create_hook_given) {
+		snapshot_creation_status = HS_READY;
+		return 0;
 	}
-	DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
-	return 1;
+	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 = HS_POST_RUNNING;
+	return ret;
 }
 
-/* exits on errors */
-void parse_config_file(int override)
+static int post_remove_hook(void)
 {
-	int ret;
+	int ret, fds[3] = {0, 0, 0};
+	char *cmd;
+	struct snapshot *s = snapshot_currently_being_removed;
+
+	assert(s);
+
+	if (!conf.post_remove_hook_given) {
+		snapshot_removal_status = HS_READY;
+		return 0;
+	}
+	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(&remove_pid, cmd, fds);
+	free(cmd);
+	if (ret < 0)
+		return ret;
+	snapshot_removal_status = HS_POST_RUNNING;
+	return ret;
+}
+
+static void kill_process(pid_t pid)
+{
+	if (!pid)
+		return;
+	DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
+	kill(pid, SIGTERM);
+}
+
+static void stop_create_process(void)
+{
+	if (!create_pid || create_process_stopped)
+		return;
+	kill(SIGSTOP, create_pid);
+	create_process_stopped = 1;
+}
+
+static void restart_create_process(void)
+{
+	if (!create_pid || !create_process_stopped)
+		return;
+	kill (SIGCONT, create_pid);
+	create_process_stopped = 0;
+}
+
+/**
+ * Print a log message about the exit status of a child.
+ */
+static void log_termination_msg(pid_t pid, int status)
+{
+	if (WIFEXITED(status))
+		DSS_INFO_LOG("child %i exited. Exit status: %i\n", (int)pid,
+			WEXITSTATUS(status));
+	else if (WIFSIGNALED(status))
+		DSS_NOTICE_LOG("child %i was killed by signal %i\n", (int)pid,
+			WTERMSIG(status));
+	else
+		DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
+}
+
+static int wait_for_process(pid_t pid, int *status)
+{
+	int ret;
+
+	DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
+	for (;;) {
+		fd_set rfds;
+
+		FD_ZERO(&rfds);
+		FD_SET(signal_pipe, &rfds);
+		ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
+		if (ret < 0)
+			break;
+		ret = next_signal();
+		if (!ret)
+			continue;
+		if (ret == SIGCHLD) {
+			ret = waitpid(pid, status, 0);
+			if (ret >= 0)
+				break;
+			if (errno != EINTR) { /* error */
+				ret = -ERRNO_TO_DSS_ERROR(errno);
+				break;
+			}
+		}
+		/* SIGINT or SIGTERM */
+		DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
+		kill(pid, SIGTERM);
+	}
+	if (ret < 0)
+		DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
+	else
+		log_termination_msg(pid, *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)
+{
+	if (!WIFEXITED(status)) {
+		snapshot_removal_status = HS_READY;
+		return -E_INVOLUNTARY_EXIT;
+	}
+	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 void handle_post_remove_exit(void)
+{
+	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_remove_exit(status);
+}
+
+static int handle_rsync_exit(int status)
+{
+	int es, ret;
+
+	if (!WIFEXITED(status)) {
+		DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
+		ret = -E_INVOLUNTARY_EXIT;
+		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)create_pid, es);
+		ret = -E_BAD_EXIT_CODE;
+		snapshot_creation_status = HS_READY;
+		goto out;
+	}
+	ret = rename_incomplete_snapshot(current_snapshot_creation_time);
+	if (ret < 0)
+		goto out;
+	snapshot_creation_status = HS_SUCCESS;
+	free(name_of_reference_snapshot);
+	name_of_reference_snapshot = NULL;
+out:
+	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 = HS_READY;
+		ret = -E_INVOLUNTARY_EXIT;
+		goto out;
+	}
+	es = WEXITSTATUS(status);
+	if (es) {
+		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 */
+		}
+		gettimeofday(&next_snapshot_time, NULL);
+		next_snapshot_time.tv_sec += 60;
+		snapshot_creation_status = HS_READY;
+		ret = 0;
+		goto out;
+	}
+	warn_count = 0;
+	snapshot_creation_status = HS_PRE_SUCCESS;
+	ret = 1;
+out:
+	create_pid = 0;
+	return ret;
+}
+
+static int handle_sigchld(void)
+{
+	pid_t pid;
+	int status, ret = reap_child(&pid, &status);
+
+	if (ret <= 0)
+		return ret;
+
+	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);
+	return -E_BUG;
+}
+
+static int check_config(void)
+{
+	if (conf.unit_interval_arg <= 0) {
+		DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
+		return -E_INVALID_NUMBER;
+	}
+	DSS_DEBUG_LOG("unit interval: %i day(s)\n", conf.unit_interval_arg);
+	if (conf.num_intervals_arg <= 0) {
+		DSS_ERROR_LOG("bad number of intervals  %i\n", conf.num_intervals_arg);
+		return -E_INVALID_NUMBER;
+	}
+	DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
+	return 1;
+}
+
+static int parse_config_file(int override)
+{
+	int ret;
 	char *config_file;
 	struct stat statbuf;
 	char *old_logfile_arg = NULL;
@@ -601,8 +831,18 @@ void parse_config_file(int override)
 			.override = override,
 			.initialize = 0,
 			.check_required = 1,
-			.check_ambiguity = 0
+			.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();
@@ -624,136 +864,85 @@ 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);
-	ret = dss_chdir(conf.dest_dir_arg);
 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;
 }
 
-void handle_sighup(void)
+static int change_to_dest_dir(void)
 {
-	DSS_NOTICE_LOG("SIGHUP\n");
-	parse_config_file(1);
+	DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
+	return dss_chdir(conf.dest_dir_arg);
 }
 
-int rename_incomplete_snapshot(int64_t start)
+static int handle_sighup(void)
 {
-	char *old_name;
 	int ret;
 
-	free(path_to_last_complete_snapshot);
-	ret = complete_name(start, get_current_time(),
-		&path_to_last_complete_snapshot);
+	DSS_NOTICE_LOG("SIGHUP\n");
+	ret = parse_config_file(1);
 	if (ret < 0)
 		return ret;
-	old_name = incomplete_name(start);
-	ret = dss_rename(old_name, path_to_last_complete_snapshot);
-	if (ret >= 0)
-		DSS_NOTICE_LOG("%s -> %s\n", old_name,
-			path_to_last_complete_snapshot);
-	free(old_name);
-	return ret;
-}
-
-void compute_next_snapshot_time(void)
-{
-	struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
-		tmp, diff;
-	int64_t x = 0;
-	unsigned wanted = num_snapshots(0), num_complete_snapshots = 0;
-	int i, ret;
-	struct snapshot *s = NULL;
-	struct snapshot_list sl;
-
-	assert(snapshot_creation_status == SCS_READY);
-	current_snapshot_creation_time = 0;
-	get_snapshot_list(&sl);
-	FOR_EACH_SNAPSHOT(s, i, &sl) {
-		if (!(s->flags & SS_COMPLETE))
-			continue;
-		num_complete_snapshots++;
-		x += s->completion_time - s->creation_time;
-	}
-	assert(x >= 0);
-	if (num_complete_snapshots)
-		x /= num_complete_snapshots; /* avg time to create one snapshot */
-	x *= wanted; /* time to create all snapshots in interval 0 */
-	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;
-	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;
-out:
-	free_snapshot_list(&sl);
+	return change_to_dest_dir();
 }
 
-int handle_rsync_exit(int status)
+static int handle_signal(void)
 {
-	int es, ret;
+	int sig, ret = next_signal();
 
-	if (!WIFEXITED(status)) {
-		DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)rsync_pid);
-		ret = -E_INVOLUNTARY_EXIT;
-		snapshot_creation_status = SCS_READY;
-		compute_next_snapshot_time();
-		goto out;
-	}
-	es = WEXITSTATUS(status);
-	if (es != 0 && es != 23 && es != 24) {
-		DSS_ERROR_LOG("rsync process %d returned %d\n", (int)rsync_pid, es);
-		ret = -E_BAD_EXIT_CODE;
-		snapshot_creation_status = SCS_READY;
-		compute_next_snapshot_time();
+	if (ret <= 0)
 		goto out;
+	sig = ret;
+	switch (sig) {
+	case SIGINT:
+	case SIGTERM:
+		restart_create_process();
+		kill_process(create_pid);
+		kill_process(remove_pid);
+		ret = -E_SIGNAL;
+		break;
+	case SIGHUP:
+		ret = handle_sighup();
+		break;
+	case SIGCHLD:
+		ret = handle_sigchld();
+		break;
 	}
-	ret = rename_incomplete_snapshot(current_snapshot_creation_time);
-	if (ret < 0)
-		goto out;
-	snapshot_creation_status = SCS_RSYNC_SUCCESS;
 out:
-	rsync_pid = 0;
-	rsync_stopped = 0;
+	if (ret < 0)
+		DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
 	return ret;
 }
 
-__malloc char *name_of_newest_complete_snapshot(void)
+/*
+ * 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)
 {
-	struct snapshot_list sl;
-	struct snapshot *s;
-	int i;
-	char *name = NULL;
-
-	get_snapshot_list(&sl);
+	char *h = conf.remote_host_arg;
 
-	FOR_EACH_SNAPSHOT_REVERSE(s, i, &sl) {
-		if (s->flags != SS_COMPLETE) /* incomplete or being deleted */
-			continue;
-		name = dss_strdup(s->name);
-		break;
-	}
-	free_snapshot_list(&sl);
-	return name;
+	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;
 }
 
-void create_rsync_argv(char ***argv, int64_t *num)
+static void create_rsync_argv(char ***argv, int64_t *num)
 {
-	char *logname, *newest = name_of_newest_complete_snapshot();
+	char *logname;
 	int i = 0, j;
+	struct snapshot_list sl;
+
+	dss_get_snapshot_list(&sl);
+	assert(!name_of_reference_snapshot);
+	name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
+	free_snapshot_list(&sl);
 
 	*argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
 	(*argv)[i++] = dss_strdup("rsync");
@@ -761,19 +950,14 @@ void create_rsync_argv(char ***argv, int64_t *num)
 	(*argv)[i++] = dss_strdup("--delete");
 	for (j = 0; j < conf.rsync_option_given; j++)
 		(*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
-	if (newest) {
-		DSS_INFO_LOG("using %s as reference snapshot\n", newest);
-		(*argv)[i++] = make_message("--link-dest=../%s", newest);
-		free(newest);
+	if (name_of_reference_snapshot) {
+		DSS_INFO_LOG("using %s as reference\n", name_of_reference_snapshot);
+		(*argv)[i++] = make_message("--link-dest=../%s",
+			name_of_reference_snapshot);
 	} else
-		DSS_INFO_LOG("no previous snapshot found\n");
-	if (conf.exclude_patterns_given) {
-		(*argv)[i++] = dss_strdup("--exclude-from");
-		(*argv)[i++] = dss_strdup(conf.exclude_patterns_arg);
-
-	}
+		DSS_INFO_LOG("no suitable reference 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?
@@ -787,53 +971,18 @@ void create_rsync_argv(char ***argv, int64_t *num)
 		DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
 }
 
-void free_rsync_argv(char **argv)
+static void free_rsync_argv(char **argv)
 {
 	int i;
+
+	if (!argv)
+		return;
 	for (i = 0; argv[i]; i++)
 		free(argv[i]);
 	free(argv);
 }
 
-int pre_create_hook(void)
-{
-	int ret, fds[3] = {0, 0, 0};
-
-	if (!conf.pre_create_hook_given) {
-		snapshot_creation_status = SCS_PRE_HOOK_SUCCESS;
-		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);
-	if (ret < 0)
-		return ret;
-	snapshot_creation_status = SCS_PRE_HOOK_RUNNING;
-	return ret;
-}
-
-int post_create_hook(void)
-{
-	int ret, fds[3] = {0, 0, 0};
-	char *cmd;
-
-	if (!conf.post_create_hook_given) {
-		snapshot_creation_status = SCS_READY;
-		compute_next_snapshot_time();
-		return 0;
-	}
-	cmd = make_message("%s %s", conf.post_create_hook_arg,
-		path_to_last_complete_snapshot);
-	DSS_NOTICE_LOG("executing %s\n", cmd);
-	ret = dss_exec_cmdline_pid(&post_create_hook_pid, cmd, fds);
-	free(cmd);
-	if (ret < 0)
-		return ret;
-	snapshot_creation_status = SCS_POST_HOOK_RUNNING;
-	return ret;
-}
-
-int create_snapshot(char **argv)
+static int create_snapshot(char **argv)
 {
 	int ret, fds[3] = {0, 0, 0};
 	char *name;
@@ -841,213 +990,114 @@ 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);
-	if (ret < 0)
-		return ret;
-	snapshot_creation_status = SCS_RSYNC_RUNNING;
-	return ret;
-}
-
-int handle_pre_create_hook_exit(int status)
-{
-	int es, ret;
-
-	if (!WIFEXITED(status)) {
-		snapshot_creation_status = SCS_READY;
-		compute_next_snapshot_time();
-		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;
-		goto out;
-	}
-	snapshot_creation_status = SCS_PRE_HOOK_SUCCESS;
-	ret = 1;
-out:
-	pre_create_hook_pid = 0;
-	return ret;
-}
-
-int handle_sigchld()
-{
-	pid_t pid;
-	int status, ret = reap_child(&pid, &status);
-
-	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;
-	}
-	DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
-	exit(EXIT_FAILURE);
-}
-
-void handle_signal(void)
-{
-	int sig, ret = next_signal();
-
-	if (ret <= 0)
-		goto out;
-	sig = ret;
-	switch (sig) {
-	case SIGINT:
-	case SIGTERM:
-		restart_rsync_process();
-		kill_process(rsync_pid);
-		kill_process(rm_pid);
-		exit(EXIT_FAILURE);
-	case SIGHUP:
-		handle_sighup();
-		ret = 1;
-		break;
-	case SIGCHLD:
-		ret = handle_sigchld();
-		break;
-	}
-out:
-	if (ret < 0)
-		DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
-}
-
-int remove_oldest_snapshot(struct snapshot_list *sl)
-{
-	struct snapshot *s = oldest_snapshot(sl);
-
-	if (!s) /* no snapshot found */
-		return 0;
-	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);
-}
-
-/* TODO: Also consider number of inodes. */
-int disk_space_low(void)
-{
-	struct disk_space ds;
-	int ret = get_disk_space(".", &ds);
-
+	ret = dss_exec(&create_pid, argv[0], argv, fds);
 	if (ret < 0)
 		return ret;
-	if (conf.min_free_mb_arg)
-		if (ds.free_mb < conf.min_free_mb_arg)
-			return 1;
-	if (conf.min_free_percent_arg)
-		if (ds.percent_free < conf.min_free_percent_arg)
-			return 1;
-	return 0;
-}
-
-int try_to_free_disk_space(int low_disk_space)
-{
-	int ret;
-	struct snapshot_list sl;
-
-	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;
-	ret = 0;
-	if (!low_disk_space)
-		goto out;
-	DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
-	ret = remove_oldest_snapshot(&sl);
-	if (ret)
-		goto out;
-	DSS_CRIT_LOG("uhuhu: not enough disk space for a single snapshot\n");
-	ret= -ENOSPC;
-out:
-	free_snapshot_list(&sl);
+	snapshot_creation_status = HS_RUNNING;
 	return ret;
 }
 
-int select_loop(void)
+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;
-		if (low_disk_space)
-			stop_rsync_process();
 		ret = try_to_free_disk_space(low_disk_space);
 		if (ret < 0)
-			break;
-		if (rm_pid)
-			continue;
-		restart_rsync_process();
-		gettimeofday(&now, NULL);
-		if (tv_diff(&next_snapshot_time, &now, &tv) > 0)
+			goto out;
+		if (snapshot_removal_status != HS_READY) {
+			stop_create_process();
 			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;
 }
 
-int com_run(void)
+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;
 
@@ -1058,57 +1108,76 @@ int com_run(void)
 	ret = install_sighandler(SIGHUP);
 	if (ret < 0)
 		return ret;
-	compute_next_snapshot_time();
-	return select_loop();
-}
-
-void log_disk_space(struct disk_space *ds)
-{
-	DSS_INFO_LOG("free: %uM/%uM (%u%%), %u%% inodes unused\n",
-		ds->free_mb, ds->total_mb, ds->percent_free,
-		ds->percent_free_inodes);
+	ret = select_loop();
+	if (ret >= 0) /* impossible */
+		ret = -E_BUG;
+	exit_hook(ret);
+	return ret;
 }
 
-int com_prune(void)
+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 (;;) {
-		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 (;;) {
-		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;
 }
 
-int com_create(void)
+static int com_create(void)
 {
 	int ret, status;
 	char **rsync_argv;
@@ -1131,45 +1200,50 @@ 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;
 }
 
-int com_ls(void)
+static int com_ls(void)
 {
 	int i;
 	struct snapshot_list sl;
 	struct snapshot *s;
-	get_snapshot_list(&sl);
-	FOR_EACH_SNAPSHOT(s, i, &sl)
-		dss_msg("%u\t%s\n", s->interval, s->name);
+
+	dss_get_snapshot_list(&sl);
+	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;
 
@@ -1177,19 +1251,19 @@ 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);
 }
 
+/**
+ * The main function of dss.
+ *
+ * \param argc Usual argument count.
+ * \param argv Usual argument vector.
+ */
 int main(int argc, char **argv)
 {
 	int ret;
@@ -1197,16 +1271,24 @@ int main(int argc, char **argv)
 		.override = 0,
 		.initialize = 1,
 		.check_required = 0,
-		.check_ambiguity = 0
+		.check_ambiguity = 0,
+		.print_errors = 1
 	};
 
 	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();
-	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);