X-Git-Url: http://git.tuebingen.mpg.de/?p=dss.git;a=blobdiff_plain;f=dss.c;h=db065c689b95dc0274231c9923cb9be6dd852839;hp=b619f7d8ffb53909d96bead3f92aff5e4a75f90e;hb=e60f419875fc41704b9e3df9914c6393e7a7791e;hpb=e4c216f105a5f8d2ce96505685395ea7d7c08dbe

diff --git a/dss.c b/dss.c
index b619f7d..db065c6 100644
--- a/dss.c
+++ b/dss.c
@@ -1,3 +1,8 @@
+/*
+ * Copyright (C) 2008-2010 Andre Noll <maan@systemlinux.org>
+ *
+ * Licensed under the GPL v2. For licencing details see COPYING.
+ */
 #include <string.h>
 #include <stdlib.h>
 #include <stdarg.h>
@@ -9,6 +14,7 @@
 #include <sys/stat.h>
 #include <unistd.h>
 #include <inttypes.h>
+#include <sys/time.h>
 #include <time.h>
 #include <sys/wait.h>
 #include <fnmatch.h>
@@ -23,14 +29,141 @@
 #include "fd.h"
 #include "exec.h"
 #include "daemon.h"
-
-
-struct gengetopt_args_info conf;
-char *dss_error_txt = NULL;
+#include "signal.h"
+#include "df.h"
+#include "time.h"
+#include "snap.h"
+
+/** 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 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 int64_t 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;
+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;
+static const char const *hook_status_description[] = {HOOK_STATUS_ARRAY};
+
+/* may be called with ds == NULL. */
+static int disk_space_low(struct disk_space *ds)
+{
+	struct disk_space ds_struct;
+
+	if (!ds) {
+		int ret = get_disk_space(".", &ds_struct);
+		if (ret < 0)
+			return ret;
+		ds = &ds_struct;
+	}
+	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;
+}
+
+static void dump_dss_config(const char *msg)
+{
+	const char dash[] = "-----------------------------";
+	int ret;
+	FILE *log = logfile? logfile : stderr;
+	struct disk_space ds;
+	int64_t now = get_current_time();
 
+	if (conf.loglevel_arg > INFO)
+		return;
+
+	fprintf(log, "%s <%s config> %s\n", dash, msg, dash);
+	fprintf(log, "\n*** disk space ***\n\n");
+	ret = get_disk_space(".", &ds);
+	if (ret >= 0) {
+		DSS_INFO_LOG("disk space low: %s\n", disk_space_low(&ds)?
+			"yes" : "no");
+		log_disk_space(&ds);
+	} else
+		DSS_ERROR_LOG("can not get free disk space: %s\n",
+			dss_strerror(-ret));
+
+	/* we continue on errors from get_disk_space */
+
+	fprintf(log, "\n*** command line and config file options ***\n\n");
+	cmdline_parser_dump(log, &conf);
+	fprintf(log, "\n*** internal state ***\n\n");
+	fprintf(log,
+		"pid: %d\n"
+		"logile: %s\n"
+		"snapshot_currently_being_removed: %s\n"
+		"path_to_last_complete_snapshot: %s\n"
+		"reference_snapshot: %s\n"
+		"snapshot_creation_status: %s\n"
+		"snapshot_removal_status: %s\n"
+		,
+		(int) getpid(),
+		logfile? conf.logfile_arg : "stderr",
+		snapshot_currently_being_removed?
+			snapshot_currently_being_removed->name : "(none)",
+		path_to_last_complete_snapshot?
+			path_to_last_complete_snapshot : "(none)",
+		name_of_reference_snapshot?
+			name_of_reference_snapshot : "(none)",
+		hook_status_description[snapshot_creation_status],
+		hook_status_description[snapshot_removal_status]
+	);
+	if (create_pid != 0)
+		fprintf(log,
+			"create_pid: %" PRId32 "\n"
+			"create process is %sstopped\n"
+			,
+			create_pid,
+			create_process_stopped? "" : "not "
+		);
+	if (remove_pid != 0)
+		fprintf(log, "remove_pid: %" PRId32 "\n", remove_pid);
+	if (next_snapshot_time != 0)
+		fprintf(log, "next snapshot due in %" PRId64 " seconds\n",
+			next_snapshot_time - now);
+	if (current_snapshot_creation_time != 0)
+		fprintf(log, "current_snapshot_creation_time: %"
+			PRId64 " (%" PRId64 " seconds ago)\n",
+			current_snapshot_creation_time,
+			now - current_snapshot_creation_time
+		);
+	if (next_removal_check.tv_sec != 0) {
+		fprintf(log, "next removal check: %llu (%llu seconds ago)\n",
+			(long long unsigned)next_removal_check.tv_sec,
+			now - (long long unsigned)next_removal_check.tv_sec
+		);
+
+	}
+	fprintf(log, "%s </%s config> %s\n", dash, msg, dash);
+}
 
 /* a litte cpp magic helps to DRY */
 #define COMMANDS \
@@ -38,38 +171,27 @@ 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 {
-	SS_COMPLETE = 1,
-	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;
@@ -92,7 +214,10 @@ __printf_2_3 void dss_log(int ll, const char* fmt,...)
 	va_end(argp);
 }
 
-__printf_1_2 void msg(const char* fmt,...)
+/**
+ * Print a message either to stdout or to the log file.
+ */
+static __printf_1_2 void dss_msg(const char* fmt,...)
 {
 	FILE *outfd = conf.daemon_given? logfile : stdout;
 	va_list argp;
@@ -101,271 +226,211 @@ __printf_1_2 void msg(const char* fmt,...)
 	va_end(argp);
 }
 
-int is_snapshot(const char *dirname, int64_t now, struct snapshot *s)
-{
-	int i, ret;
-	char *dash, *dot, *tmp;
-	int64_t num;
-
-	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)
-		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;
-	}
-	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;
-	}
-	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;
-}
-
-char *incomplete_name(int64_t start)
-{
-	return make_message("%lli-incomplete", (long long)start);
-}
-
-char *being_deleted_name(struct snapshot *s)
+static void dss_get_snapshot_list(struct snapshot_list *sl)
 {
-	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);
+	get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
 }
 
-int complete_name(int64_t start, int64_t end, char **result)
+static int64_t compute_next_snapshot_time(void)
 {
-	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];
+	int64_t x = 0, now = get_current_time(), unit_interval
+		= 24 * 3600 * conf.unit_interval_arg, ret;
+	unsigned wanted = desired_number_of_snapshots(0, conf.num_intervals_arg),
+		num_complete_snapshots = 0;
+	int i;
+	struct snapshot *s = NULL;
+	struct snapshot_list sl;
 
-	if (!localtime_r(start_seconds, &start_tm)) {
-		make_err_msg("%lli", (long long)start);
-		return -E_LOCALTIME;
-	}
-	if (!localtime_r(end_seconds, &end_tm)) {
-		make_err_msg("%lli", (long long)end);
-		return -E_LOCALTIME;
-	}
-	if (!strftime(start_str, sizeof(start_str), "%a_%b_%d_%Y_%H_%M_%S", &start_tm)) {
-		make_err_msg("%lli", (long long)start);
-		return -E_STRFTIME;
-	}
-	if (!strftime(end_str, sizeof(end_str), "%a_%b_%d_%Y_%H_%M_%S", &end_tm)) {
-		make_err_msg("%lli", (long long)end);
-		return -E_STRFTIME;
+	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;
 	}
-	*result = make_message("%lli-%lli.%s-%s", (long long) start, (long long) end,
-		start_str, end_str);
-	return 1;
-}
+	assert(x >= 0);
 
-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;
-};
-
-#define FOR_EACH_SNAPSHOT(s, i, sl) \
-	for ((i) = 0; (i) < (sl)->num_snapshots && ((s) = (sl)->snapshots[(i)]); (i)++)
-
-
-
-#define NUM_COMPARE(x, y) ((int)((x) < (y)) - (int)((x) > (y)))
-
-static int compare_snapshots(const void *a, const void *b)
-{
-	struct snapshot *s1 = *(struct snapshot **)a;
-	struct snapshot *s2 = *(struct snapshot **)b;
-	return NUM_COMPARE(s2->creation_time, s1->creation_time);
+	ret = now;
+	if (num_complete_snapshots == 0)
+		goto out;
+	x /= num_complete_snapshots; /* avg time to create one snapshot */
+	if (unit_interval < x * wanted) /* oops, no sleep at all */
+		goto out;
+	ret = s->completion_time + unit_interval / wanted - x;
+out:
+	free_snapshot_list(&sl);
+	return ret;
 }
 
-/** Compute the minimum of \a a and \a b. */
-#define DSS_MIN(a,b) ((a) < (b) ? (a) : (b))
-
-int add_snapshot(const char *dirname, void *private)
+static inline void invalidate_next_snapshot_time(void)
 {
-	struct snapshot_list *sl = private;
-	struct snapshot s;
-	int ret = is_snapshot(dirname, sl->now, &s);
-
-	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;
+	next_snapshot_time = 0;
 }
 
-void get_snapshot_list(struct snapshot_list *sl)
+static inline int next_snapshot_time_is_valid(void)
 {
-	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 next_snapshot_time != 0;
 }
 
-void free_snapshot_list(struct snapshot_list *sl)
+static int next_snapshot_is_due(void)
 {
-	int i;
-	struct snapshot *s;
+	int64_t now = get_current_time();
 
-	FOR_EACH_SNAPSHOT(s, i, sl) {
-		free(s->name);
-		free(s);
+	if (!next_snapshot_time_is_valid())
+		next_snapshot_time = compute_next_snapshot_time();
+	if (next_snapshot_time <= now) {
+		DSS_DEBUG_LOG("next snapshot: now\n");
+		return 1;
 	}
-	free(sl->interval_count);
-	free(sl->snapshots);
+	DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n",
+		next_snapshot_time - now);
+	return 0;
 }
 
-/**
- * Print a log message about the exit status of a child.
- */
-void log_termination_msg(pid_t pid, int status)
+static int pre_create_hook(void)
 {
-	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 ret, fds[3] = {0, 0, 0};
+
+	assert(snapshot_creation_status == HS_READY);
+	/* make sure that the next snapshot time will be recomputed */
+	invalidate_next_snapshot_time();
+	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 wait_for_process(pid_t pid, int *status)
+static int pre_remove_hook(struct snapshot *s, const char *why)
 {
-	int ret;
+	int ret, fds[3] = {0, 0, 0};
+	char *cmd;
 
-	DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
-	for (;;) {
-		ret = waitpid(pid, status, 0);
-		if (ret >= 0 || errno != EINTR)
-			break;
-	}
-	if (ret < 0) {
-		ret = -ERRNO_TO_DSS_ERROR(errno);
-		make_err_msg("failed to wait for process %d", (int)pid);
-	} else
-		log_termination_msg(pid, *status);
+	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);
+
+	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, cmd, fds);
+	free(cmd);
+	if (ret < 0)
+		return ret;
+	snapshot_removal_status = HS_PRE_RUNNING;
 	return ret;
 }
 
-int remove_snapshot(struct snapshot *s, pid_t *pid)
+static int exec_rm(void)
 {
+	struct snapshot *s = snapshot_currently_being_removed;
 	int fds[3] = {0, 0, 0};
 	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(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;
 }
 
-int remove_redundant_snapshot(struct snapshot_list *sl,
-		int dry_run, pid_t *pid)
+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;
+}
+
+static int is_reference_snapshot(struct snapshot *s)
+{
+	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
+ */
+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 = 1<<(conf.num_intervals_arg - interval - 1);
+		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;
 
 		if (keep >= num)
 			missing += keep - num;
-		DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
-			interval, keep, num, missing);
+//		DSS_DEBUG_LOG("interval %i: keep: %u, have: %u, missing: %u\n",
+//			interval, keep, num, missing);
 		if (keep + missing >= num)
 			continue;
 		/* redundant snapshot in this interval, pick snapshot with lowest score */
 		FOR_EACH_SNAPSHOT(s, i, sl) {
 			int64_t this_score;
 
-			DSS_DEBUG_LOG("checking %s\n", s->name);
+			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;
 				continue;
@@ -381,7 +446,7 @@ int remove_redundant_snapshot(struct snapshot_list *sl,
 			/* check if s is a better victim */
 			this_score = s->creation_time - prev->creation_time;
 			assert(this_score >= 0);
-			DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
+			//DSS_DEBUG_LOG("%s: score %lli\n", s->name, (long long)score);
 			if (this_score < score) {
 				score = this_score;
 				victim = s;
@@ -389,257 +454,462 @@ int remove_redundant_snapshot(struct snapshot_list *sl,
 			prev = s;
 		}
 		assert(victim);
-		if (dry_run) {
-			msg("%s would be removed (interval = %i)\n",
-				victim->name, victim->interval);
-			continue;
-		}
-		ret = remove_snapshot(victim, pid);
-		return ret < 0? ret : 1;
+		return victim;
 	}
-	return 0;
+	return NULL;
 }
 
-int remove_old_snapshot(struct snapshot_list *sl, int dry_run, pid_t *pid)
+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 >= %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 (dry_run) {
-			msg("%s would be removed (interval = %i)\n",
-				s->name, s->interval);
+		if (is_reference_snapshot(s))
 			continue;
-		}
-		ret = remove_snapshot(s, pid);
-		if (ret < 0)
-			return ret;
-		return 1;
+		if (s->interval < conf.num_intervals_arg)
+			continue;
+		return s;
 	}
-	return 0;
+	return NULL;
 }
 
-int wait_for_rm_process(pid_t pid)
+struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 {
-	int status, es, ret = wait_for_process(pid, &status);
-	if (ret < 0)
-		return ret;
-	if (!WIFEXITED(status)) {
-		ret = E_INVOLUNTARY_EXIT;
-		make_err_msg("rm process %d died involuntary", (int)pid);
-		return ret;
-	}
-	es = WEXITSTATUS(status);
-	if (es) {
-		ret = -E_BAD_EXIT_CODE;
-		make_err_msg("rm process %d returned %d", (int)pid, es);
-		return ret;
+	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 1;
+	return NULL;
 }
 
-int com_run(void)
+static int rename_incomplete_snapshot(int64_t start)
 {
-	return 42;
+	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;
+	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;
 }
 
-int com_prune(void)
+static int try_to_free_disk_space(void)
 {
-	int ret, dry_run = 0;
+	int ret;
 	struct snapshot_list sl;
-	pid_t pid;
+	struct snapshot *victim;
+	struct timeval now;
+	const char *why;
+	int low_disk_space;
 
-	for (;;) {
-		get_snapshot_list(&sl);
-		ret = remove_old_snapshot(&sl, dry_run, &pid);
-		free_snapshot_list(&sl);
-		if (ret < 0)
-			return ret;
-		if (!ret)
-			break;
-		ret = wait_for_rm_process(pid);
-		if (ret < 0)
-			goto out;
-	}
-	for (;;) {
-		get_snapshot_list(&sl);
-		ret = remove_redundant_snapshot(&sl, dry_run, &pid);
-		free_snapshot_list(&sl);
-		if (ret < 0)
-			return ret;
-		if (!ret)
-			break;
-		ret = wait_for_rm_process(pid);
-		if (ret < 0)
-			goto out;
+	ret = disk_space_low(NULL);
+	if (ret < 0)
+		return ret;
+	low_disk_space = ret;
+	gettimeofday(&now, NULL);
+	if (tv_diff(&next_removal_check, &now, NULL) > 0)
+		return 0;
+	if (!low_disk_space) {
+		if (conf.keep_redundant_given)
+			return 0;
+		if (snapshot_creation_status != HS_READY)
+			return 0;
+		if (next_snapshot_is_due())
+			return 0;
 	}
-	return 1;
+	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;
 }
 
-struct newest_snapshot_data {
-	char * newest_name;
-	int64_t newest_creation_time;
-	int64_t now;
-};
-
-int get_newest_complete(const char *dirname, void *private)
+static int post_create_hook(void)
 {
-	struct newest_snapshot_data *nsd = private;
-	struct snapshot s;
-	int ret = is_snapshot(dirname, nsd->now, &s);
-
-	if (ret <= 0)
-		return 1;
-	if (s.creation_time < nsd->newest_creation_time)
-		return 1;
-	nsd->newest_creation_time = s.creation_time;
-	free(nsd->newest_name);
-	nsd->newest_name = s.name;
-	return 1;
+	int ret, fds[3] = {0, 0, 0};
+	char *cmd;
+
+	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;
 }
 
-__malloc char *name_of_newest_complete_snapshot(void)
+static int post_remove_hook(void)
 {
-	struct newest_snapshot_data nsd = {
-		.now = get_current_time(),
-		.newest_creation_time = -1
-	};
-	for_each_subdir(get_newest_complete, &nsd);
-	return nsd.newest_name;
+	int ret, fds[3] = {0, 0, 0};
+	char *cmd;
+	struct snapshot *s = snapshot_currently_being_removed;
+
+	assert(s);
+
+	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;
 }
 
-void create_rsync_argv(char ***argv, int64_t *num)
+static void dss_kill(pid_t pid, int sig, const char *msg)
 {
-	char *logname, *newest = name_of_newest_complete_snapshot();
-	int i = 0, j;
-
-	*argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
-	(*argv)[i++] = dss_strdup("rsync");
-	(*argv)[i++] = dss_strdup("-aq");
-	(*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);
-	} else
-		DSS_INFO_LOG("no previous snapshot found");
-	if (conf.exclude_patterns_given) {
-		(*argv)[i++] = dss_strdup("--exclude-from");
-		(*argv)[i++] = dss_strdup(conf.exclude_patterns_arg);
+	const char *signame, *process_name;
 
+	if (pid == 0)
+		return;
+	switch (sig) {
+	case SIGTERM: signame = "TERM"; break;
+	case SIGSTOP: signame = "STOP"; break;
+	case SIGCONT: signame = "CONT"; break;
+	default: signame = "????";
 	}
-	logname = dss_logname();
-	if (conf.remote_user_given && !strcmp(conf.remote_user_arg, logname))
-		(*argv)[i++] = dss_strdup(conf.source_dir_arg);
-	else
-		(*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
-			conf.remote_user_arg : logname,
-			conf.remote_host_arg, conf.source_dir_arg);
-	free(logname);
-	*num = get_current_time();
-	(*argv)[i++] = incomplete_name(*num);
-	(*argv)[i++] = NULL;
-	for (j = 0; j < i; j++)
-		DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
+
+	if (pid == create_pid)
+		process_name = "create";
+	else if (pid == remove_pid)
+		process_name = "remove";
+	else process_name = "??????";
+
+	if (msg)
+		DSS_INFO_LOG("%s\n", msg);
+	DSS_DEBUG_LOG("sending signal %d (%s) to pid %d (%s process)\n",
+		sig, signame, (int)pid, process_name);
+	if (kill(pid, sig) >= 0)
+		return;
+	DSS_INFO_LOG("failed to send signal %d (%s) to pid %d (%s process)\n",
+		sig, signame, (int)pid, process_name);
 }
 
-void free_rsync_argv(char **argv)
+static void stop_create_process(void)
 {
-	int i;
-	for (i = 0; argv[i]; i++)
-		free(argv[i]);
-	free(argv);
+	if (create_process_stopped)
+		return;
+	dss_kill(create_pid, SIGSTOP, "suspending create process");
+	create_process_stopped = 1;
 }
 
-int create_snapshot(char **argv, pid_t *pid)
+static void restart_create_process(void)
 {
-	int fds[3] = {0, 0, 0};
+	if (!create_process_stopped)
+		return;
+	dss_kill(create_pid, SIGCONT, "resuming create process");
+	create_process_stopped = 0;
+}
 
-	return dss_exec(pid, argv[0], argv, fds);
+/**
+ * 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);
 }
 
-int rename_incomplete_snapshot(int64_t start)
+static int wait_for_process(pid_t pid, int *status)
 {
-	char *old_name, *new_name;
 	int ret;
 
-	ret = complete_name(start, get_current_time(), &new_name);
-	if (ret < 0)
-		return ret;
-	old_name = incomplete_name(start);
-	ret = dss_rename(old_name, new_name);
-	if (ret >= 0)
-		DSS_NOTICE_LOG("%s -> %s\n", old_name, new_name);
-	free(old_name);
-	free(new_name);
+	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_kill(pid, SIGTERM, "killing child process");
+	}
+	if (ret < 0)
+		DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
+	else
+		log_termination_msg(pid, *status);
 	return ret;
 }
 
-int com_create(void)
+static void handle_pre_remove_exit(int status)
 {
-	int ret, status, es;
-	char **rsync_argv;
-	int64_t snapshot_num;
-	pid_t pid;
+	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;
+	}
+	snapshot_removal_status = HS_SUCCESS;
+	return 1;
+}
 
-	create_rsync_argv(&rsync_argv, &snapshot_num);
-	DSS_NOTICE_LOG("creating snapshot %lli\n", (long long)snapshot_num);
-	ret = create_snapshot(rsync_argv, &pid);
+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;
-	ret = wait_for_process(pid, &status);
+	}
+	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;
+		next_snapshot_time = get_current_time() + 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_process_stopped = 0;
+	return ret;
+}
+
+static int handle_pre_create_hook_exit(int status)
+{
+	int es, ret;
+	static int warn_count;
+
 	if (!WIFEXITED(status)) {
-		ret = E_INVOLUNTARY_EXIT;
-		make_err_msg("rsync process %d died involuntary", (int)pid);
+		snapshot_creation_status = HS_READY;
+		ret = -E_INVOLUNTARY_EXIT;
 		goto out;
 	}
 	es = WEXITSTATUS(status);
-	if (es != 0 && es != 23 && es != 24) {
-		ret = -E_BAD_EXIT_CODE;
-		make_err_msg("rsync process %d returned %d", (int)pid, es);
+	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 */
+		}
+		next_snapshot_time = get_current_time() + 60;
+		snapshot_creation_status = HS_READY;
+		ret = 0;
 		goto out;
 	}
-	ret = rename_incomplete_snapshot(snapshot_num);
+	warn_count = 0;
+	snapshot_creation_status = HS_PRE_SUCCESS;
+	ret = 1;
 out:
-	free_rsync_argv(rsync_argv);
 	return ret;
 }
 
-int com_ls(void)
+static int handle_sigchld(void)
 {
-	int i;
-	struct snapshot_list sl;
-	struct snapshot *s;
-	get_snapshot_list(&sl);
-	FOR_EACH_SNAPSHOT(s, i, &sl)
-		msg("%u\t%s\n", s->interval, s->name);
-	free_snapshot_list(&sl);
-	return 1;
+	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:
+			ret = handle_pre_create_hook_exit(status);
+			break;
+		case HS_RUNNING:
+			ret = handle_rsync_exit(status);
+			break;
+		case HS_POST_RUNNING:
+			snapshot_creation_status = HS_READY;
+			ret = 1;
+			break;
+		default:
+			DSS_EMERG_LOG("BUG: create can't die in status %d\n",
+				snapshot_creation_status);
+			return -E_BUG;
+		}
+		create_pid = 0;
+		return ret;
+	}
+	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;
 }
 
-/* TODO: Unlink pid file */
-__noreturn void clean_exit(int status)
+static int check_config(void)
 {
-	//kill(0, SIGTERM);
-	free(dss_error_txt);
-	exit(status);
+	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;
 }
 
-int read_config_file(void)
+/*
+ * Returns < 0 on errors, 0 if no config file is given and > 0 if the config
+ * file was read successfully.
+ */
+static int parse_config_file(int override)
 {
-	int ret;
+	int ret, config_file_exists;
 	char *config_file;
 	struct stat statbuf;
+	char *old_logfile_arg = NULL;
+	int old_daemon_given = 0;
 
 	if (conf.config_file_given)
 		config_file = dss_strdup(conf.config_file_arg);
@@ -648,75 +918,524 @@ int read_config_file(void)
 		config_file = make_message("%s/.dssrc", home);
 		free(home);
 	}
-	ret = stat(config_file, &statbuf);
-	if (ret && conf.config_file_given) {
+	if (override) { /* SIGHUP */
+		if (conf.logfile_given)
+			old_logfile_arg = dss_strdup(conf.logfile_arg);
+		old_daemon_given = conf.daemon_given;
+	}
+
+	config_file_exists = !stat(config_file, &statbuf);
+	if (!config_file_exists && conf.config_file_given) {
 		ret = -ERRNO_TO_DSS_ERROR(errno);
-		make_err_msg("failed to stat config file %s", config_file);
+		DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
 		goto out;
 	}
-	if (!ret) {
+	if (config_file_exists) {
 		struct cmdline_parser_params params = {
-			.override = 0,
+			.override = override,
 			.initialize = 0,
-			.check_required = 0,
-			.check_ambiguity = 0
+			.check_required = 1,
+			.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);
 	}
-	if (!conf.source_dir_given || !conf.dest_dir_given) {
-		ret = -E_SYNTAX;
-		make_err_msg("you need to specify both source_dir and dest_dir");
+	ret = check_config();
+	if (ret < 0)
 		goto out;
+	if (override) {
+		/* don't change daemon mode on SIGHUP */
+		conf.daemon_given = old_daemon_given;
+		close_log(logfile);
+		logfile = NULL;
+		if (conf.logfile_given)
+			free(old_logfile_arg);
+		else if (conf.daemon_given) { /* re-use old logfile */
+			conf.logfile_arg = old_logfile_arg;
+			conf.logfile_given = 1;
+		}
 	}
-	ret = 1;
+	if (conf.logfile_given) {
+		logfile = open_log(conf.logfile_arg);
+		log_welcome(conf.loglevel_arg);
+	}
+	DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
+	ret = config_file_exists;
 out:
 	free(config_file);
+	if (ret < 0)
+		DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
 	return ret;
 }
 
-int check_config(void)
+static int change_to_dest_dir(void)
 {
-	if (conf.unit_interval_arg <= 0) {
-		make_err_msg("bad unit interval: %i", 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) {
-		make_err_msg("bad number of intervals  %i", conf.num_intervals_arg);
-		return -E_INVALID_NUMBER;
+	DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
+	return dss_chdir(conf.dest_dir_arg);
+}
+
+static int handle_sighup(void)
+{
+	int ret;
+
+	DSS_NOTICE_LOG("SIGHUP, re-reading config\n");
+	dump_dss_config("old");
+	ret = parse_config_file(1);
+	if (ret < 0)
+		return ret;
+	dump_dss_config("reloaded");
+	invalidate_next_snapshot_time();
+	return change_to_dest_dir();
+}
+
+static int handle_signal(void)
+{
+	int sig, ret = next_signal();
+
+	if (ret <= 0)
+		goto out;
+	sig = ret;
+	switch (sig) {
+	case SIGINT:
+	case SIGTERM:
+		restart_create_process();
+		dss_kill(create_pid, SIGTERM, NULL);
+		dss_kill(remove_pid, SIGTERM, NULL);
+		ret = -E_SIGNAL;
+		break;
+	case SIGHUP:
+		ret = handle_sighup();
+		break;
+	case SIGCHLD:
+		ret = handle_sigchld();
+		break;
 	}
-	DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
+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;
 }
 
-int main(int argc, char **argv)
+static int rename_resume_snap(int64_t creation_time)
 {
+	struct snapshot_list sl = {.num_snapshots = 0};
+	struct snapshot *s;
+	char *new_name = incomplete_name(creation_time);
 	int ret;
 
-	cmdline_parser(argc, argv, &conf); /* aborts on errors */
-	if (conf.inputs_num) {
-		ret = -E_SYNTAX;
-		make_err_msg("additional non-options given");
+	ret = 0;
+	if (conf.no_resume_given)
+		goto out;
+	dss_get_snapshot_list(&sl);
+	s = get_newest_snapshot(&sl);
+	if (!s)
+		goto out;
+	if ((s->flags & SS_COMPLETE) != 0) /* complete */
+		goto out;
+	DSS_INFO_LOG("resuming: reusing %s as destination dir\n", s->name);
+	ret = dss_rename(s->name, new_name);
+out:
+	if (ret >= 0)
+		DSS_NOTICE_LOG("creating new snapshot %s\n", new_name);
+	free(new_name);
+	free_snapshot_list(&sl);
+	return ret;
+}
+
+static void create_rsync_argv(char ***argv, int64_t *num)
+{
+	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");
+	(*argv)[i++] = dss_strdup("-aq");
+	(*argv)[i++] = dss_strdup("--delete");
+	for (j = 0; j < conf.rsync_option_given; j++)
+		(*argv)[i++] = dss_strdup(conf.rsync_option_arg[j]);
+	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 suitable reference snapshot found\n");
+	logname = dss_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?
+			conf.remote_user_arg : logname,
+			conf.remote_host_arg, conf.source_dir_arg);
+	free(logname);
+	*num = get_current_time();
+	(*argv)[i++] = incomplete_name(*num);
+	(*argv)[i++] = NULL;
+	for (j = 0; j < i; j++)
+		DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
+}
+
+static void free_rsync_argv(char **argv)
+{
+	int i;
+
+	if (!argv)
+		return;
+	for (i = 0; argv[i]; i++)
+		free(argv[i]);
+	free(argv);
+}
+
+static int create_snapshot(char **argv)
+{
+	int ret, fds[3] = {0, 0, 0};
+
+	ret = rename_resume_snap(current_snapshot_creation_time);
+	if (ret < 0)
+		return ret;
+	ret = dss_exec(&create_pid, argv[0], argv, fds);
+	if (ret < 0)
+		return ret;
+	snapshot_creation_status = HS_RUNNING;
+	return ret;
+}
+
+static int select_loop(void)
+{
+	int ret;
+	/* check every 60 seconds for free disk space */
+	struct timeval tv;
+	char **rsync_argv = NULL;
+
+	for (;;) {
+		fd_set rfds;
+		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);
+		ret = dss_select(signal_pipe + 1, &rfds, NULL, tvp);
+		if (ret < 0)
+			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 = try_to_free_disk_space();
+		if (ret < 0)
+			goto out;
+		if (snapshot_removal_status != HS_READY) {
+			stop_create_process();
+			continue;
+		}
+		restart_create_process();
+		switch (snapshot_creation_status) {
+		case HS_READY:
+			if (!next_snapshot_is_due())
+				continue;
+			ret = pre_create_hook();
+			if (ret < 0)
+				goto out;
+			continue;
+		case HS_PRE_RUNNING:
+		case HS_RUNNING:
+		case HS_POST_RUNNING:
+			continue;
+		case HS_PRE_SUCCESS:
+			if (!name_of_reference_snapshot) {
+				free_rsync_argv(rsync_argv);
+				create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
+			}
+			ret = create_snapshot(rsync_argv);
+			if (ret < 0)
+				goto out;
+			continue;
+		case HS_NEEDS_RESTART:
+			if (!next_snapshot_is_due())
+				continue;
+			ret = create_snapshot(rsync_argv);
+			if (ret < 0)
+				goto out;
+			continue;
+		case HS_SUCCESS:
+			ret = post_create_hook();
+			if (ret < 0)
+				goto out;
+			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;
+
+	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;
+
+	if (conf.dry_run_given) {
+		DSS_ERROR_LOG("dry_run not supported by this command\n");
+		return -E_SYNTAX;
+	}
+	ret = install_sighandler(SIGHUP);
+	if (ret < 0)
+		return ret;
+	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);
+	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;
 	}
-	ret = read_config_file();
+	ret = pre_remove_hook(victim, why);
 	if (ret < 0)
 		goto out;
-	ret = check_config();
+	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;
+	}
+	ret = exec_rm();
 	if (ret < 0)
 		goto out;
-	if (conf.logfile_given) {
-		logfile = open_log(conf.logfile_arg);
-		log_welcome(conf.loglevel_arg);
+	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;
+}
+
+static int com_create(void)
+{
+	int ret, status;
+	char **rsync_argv;
+
+	if (conf.dry_run_given) {
+		int i;
+		char *msg = NULL;
+		create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
+		for (i = 0; rsync_argv[i]; i++) {
+			char *tmp = msg;
+			msg = make_message("%s%s%s", tmp? tmp : "",
+				tmp? " " : "", rsync_argv[i]);
+			free(tmp);
+		}
+		free_rsync_argv(rsync_argv);
+		dss_msg("%s\n", msg);
+		free(msg);
+		return 1;
+	}
+	ret = pre_create_hook();
+	if (ret < 0)
+		return ret;
+	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) /* 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(create_pid, &status);
+	if (ret < 0)
+		goto out;
+	ret = handle_rsync_exit(status);
+	if (ret < 0)
+		goto out;
+	post_create_hook();
+	if (create_pid)
+		ret = wait_for_process(create_pid, &status);
+out:
+	free_rsync_argv(rsync_argv);
+	return ret;
+}
+
+static int com_ls(void)
+{
+	int i;
+	struct snapshot_list sl;
+	struct snapshot *s;
+
+	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 int setup_signal_handling(void)
+{
+	int ret;
+
+	DSS_INFO_LOG("setting up signal handlers\n");
+	signal_pipe = signal_init(); /* always successful */
+	ret = install_sighandler(SIGINT);
+	if (ret < 0)
+		return ret;
+	ret = install_sighandler(SIGTERM);
+	if (ret < 0)
+		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;
+	struct cmdline_parser_params params = {
+		.override = 0,
+		.initialize = 1,
+		.check_required = 0,
+		.check_ambiguity = 0,
+		.print_errors = 1
+	};
+
+	cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
+	ret = parse_config_file(0);
+	if (ret < 0)
+		goto out;
+	if (ret == 0) { /* no config file given */
+		/*
+		 * Parse the command line options again, but this time check
+		 * that all required options are given.
+		 */
+		params = (struct cmdline_parser_params) {
+			.override = 1,
+			.initialize = 1,
+			.check_required = 1,
+			.check_ambiguity = 1,
+			.print_errors = 1
+		};
+		cmdline_parser_ext(argc, argv, &conf, &params); /* aborts on errors */
 	}
 	if (conf.daemon_given)
 		daemon_init();
-	ret = dss_chdir(conf.dest_dir_arg);
+	ret = change_to_dest_dir();
+	if (ret < 0)
+		goto out;
+	dump_dss_config("startup");
+	ret = setup_signal_handling();
 	if (ret < 0)
 		goto out;
 	ret = call_command_handler();
 out:
 	if (ret < 0)
-		log_err_msg(EMERG, -ret);
-	clean_exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
+		DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
+	exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
 }