X-Git-Url: http://git.tuebingen.mpg.de/?p=dss.git;a=blobdiff_plain;f=dss.c;h=7c8b54fc6a0502d0a2ab983b5bc8ff3d33e0bda2;hp=68efc2b07cc457c4944013b6e3147fdc3f2adce1;hb=9528cf1dad242f2f41757fb1c7cb93acdba2370d;hpb=d28be18d969b6e72021e7229be49c02d3b549794

diff --git a/dss.c b/dss.c
index 68efc2b..7c8b54f 100644
--- a/dss.c
+++ b/dss.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2008-2009 Andre Noll <maan@systemlinux.org>
+ * Copyright (C) 2008-2011 Andre Noll <maan@systemlinux.org>
  *
  * Licensed under the GPL v2. For licencing details see COPYING.
  */
@@ -64,7 +64,106 @@ enum hook_status snapshot_removal_status;
 
 
 DEFINE_DSS_ERRLIST;
+static const char *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 \
@@ -127,25 +226,6 @@ static __printf_1_2 void dss_msg(const char* fmt,...)
 	va_end(argp);
 }
 
-static int disk_space_low(void)
-{
-	struct disk_space ds;
-	int ret = get_disk_space(".", &ds);
-
-	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;
-}
-
 static void dss_get_snapshot_list(struct snapshot_list *sl)
 {
 	get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
@@ -207,33 +287,22 @@ static int next_snapshot_is_due(void)
 	return 0;
 }
 
-static int pre_create_hook(void)
+static void pre_create_hook(void)
 {
-	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();
-	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;
+	dss_exec_cmdline_pid(&create_pid, conf.pre_create_hook_arg);
 	snapshot_creation_status = HS_PRE_RUNNING;
-	return ret;
 }
 
-static int pre_remove_hook(struct snapshot *s, const char *why)
+static void pre_remove_hook(struct snapshot *s, const char *why)
 {
-	int ret, fds[3] = {0, 0, 0};
 	char *cmd;
 
 	if (!s)
-		return 0;
+		return;
 	DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
 	assert(snapshot_removal_status == HS_READY);
 	assert(remove_pid == 0);
@@ -243,30 +312,26 @@ static int pre_remove_hook(struct snapshot *s, const char *why)
 	*snapshot_currently_being_removed = *s;
 	snapshot_currently_being_removed->name = dss_strdup(s->name);
 
-	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);
+	dss_exec_cmdline_pid(&remove_pid, cmd);
 	free(cmd);
-	if (ret < 0)
-		return ret;
 	snapshot_removal_status = HS_PRE_RUNNING;
-	return ret;
 }
 
 static int exec_rm(void)
 {
 	struct snapshot *s = snapshot_currently_being_removed;
-	int fds[3] = {0, 0, 0};
 	char *new_name = being_deleted_name(s);
-	char *argv[] = {"rm", "-rf", new_name, NULL};
+	char *argv[4];
 	int ret;
 
+	argv[0] = "rm";
+	argv[1] = "-rf";
+	argv[2] = new_name;
+	argv[3] = NULL;
+
 	assert(snapshot_removal_status == HS_PRE_SUCCESS);
 	assert(remove_pid == 0);
 
@@ -274,9 +339,7 @@ static int exec_rm(void)
 	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;
+	dss_exec(&remove_pid, argv[0], argv);
 	snapshot_removal_status = HS_RUNNING;
 out:
 	free(new_name);
@@ -450,7 +513,7 @@ static int try_to_free_disk_space(void)
 	const char *why;
 	int low_disk_space;
 
-	ret = disk_space_low();
+	ret = disk_space_low(NULL);
 	if (ret < 0)
 		return ret;
 	low_disk_space = ret;
@@ -492,77 +555,79 @@ static int try_to_free_disk_space(void)
 	ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
 	goto out;
 remove:
-	ret = pre_remove_hook(victim, why);
+	pre_remove_hook(victim, why);
 out:
 	free_snapshot_list(&sl);
 	return ret;
 }
 
-static int post_create_hook(void)
+static void post_create_hook(void)
 {
-	int ret, fds[3] = {0, 0, 0};
-	char *cmd;
-
-	if (!conf.post_create_hook_given) {
-		create_pid = 0;
-		snapshot_creation_status = HS_READY;
-		return 0;
-	}
-	cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
+	char *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);
+	dss_exec_cmdline_pid(&create_pid, cmd);
 	free(cmd);
-	if (ret < 0)
-		return ret;
 	snapshot_creation_status = HS_POST_RUNNING;
-	return ret;
 }
 
-static int post_remove_hook(void)
+static void post_remove_hook(void)
 {
-	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);
+	dss_exec_cmdline_pid(&remove_pid, cmd);
 	free(cmd);
-	if (ret < 0)
-		return ret;
 	snapshot_removal_status = HS_POST_RUNNING;
-	return ret;
 }
 
-static void kill_process(pid_t pid)
+static void dss_kill(pid_t pid, int sig, const char *msg)
 {
-	if (!pid)
+	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 = "????";
+	}
+
+	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_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
-	kill(pid, SIGTERM);
+	DSS_INFO_LOG("failed to send signal %d (%s) to pid %d (%s process)\n",
+		sig, signame, (int)pid, process_name);
 }
 
 static void stop_create_process(void)
 {
-	if (!create_pid || create_process_stopped)
+	if (create_process_stopped)
 		return;
-	kill(SIGSTOP, create_pid);
+	dss_kill(create_pid, SIGSTOP, "suspending create process");
 	create_process_stopped = 1;
 }
 
 static void restart_create_process(void)
 {
-	if (!create_pid || !create_process_stopped)
+	if (!create_process_stopped)
 		return;
-	kill (SIGCONT, create_pid);
+	dss_kill(create_pid, SIGCONT, "resuming create process");
 	create_process_stopped = 0;
 }
 
@@ -607,8 +672,7 @@ static int wait_for_process(pid_t pid, int *status)
 			}
 		}
 		/* SIGINT or SIGTERM */
-		DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
-		kill(pid, SIGTERM);
+		dss_kill(pid, SIGTERM, "killing child process");
 	}
 	if (ret < 0)
 		DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
@@ -638,10 +702,7 @@ static int handle_rm_exit(int 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;
+	snapshot_removal_status = HS_SUCCESS;
 	return 1;
 }
 
@@ -778,18 +839,22 @@ static int handle_sigchld(void)
 	if (pid == create_pid) {
 		switch (snapshot_creation_status) {
 		case HS_PRE_RUNNING:
-			return handle_pre_create_hook_exit(status);
+			ret = handle_pre_create_hook_exit(status);
+			break;
 		case HS_RUNNING:
-			return handle_rsync_exit(status);
+			ret = handle_rsync_exit(status);
+			break;
 		case HS_POST_RUNNING:
 			snapshot_creation_status = HS_READY;
-			return 1;
+			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);
@@ -881,7 +946,7 @@ static int parse_config_file(int override)
 			conf.logfile_given = 1;
 		}
 	}
-	if (conf.logfile_given) {
+	if (conf.logfile_given && conf.run_given && conf.daemon_given) {
 		logfile = open_log(conf.logfile_arg);
 		log_welcome(conf.loglevel_arg);
 	}
@@ -900,24 +965,16 @@ static int change_to_dest_dir(void)
 	return dss_chdir(conf.dest_dir_arg);
 }
 
-static void dump_dss_config(const char *msg)
-{
-	if (conf.loglevel_arg > INFO)
-		return;
-	DSS_INFO_LOG("%s\n", msg);
-	cmdline_parser_dump(logfile? logfile : stderr, &conf);
-}
-
 static int handle_sighup(void)
 {
 	int ret;
 
 	DSS_NOTICE_LOG("SIGHUP, re-reading config\n");
-	dump_dss_config("current config");
+	dump_dss_config("old");
 	ret = parse_config_file(1);
 	if (ret < 0)
 		return ret;
-	dump_dss_config("new config");
+	dump_dss_config("reloaded");
 	invalidate_next_snapshot_time();
 	return change_to_dest_dir();
 }
@@ -933,8 +990,8 @@ static int handle_signal(void)
 	case SIGINT:
 	case SIGTERM:
 		restart_create_process();
-		kill_process(create_pid);
-		kill_process(remove_pid);
+		dss_kill(create_pid, SIGTERM, NULL);
+		dss_kill(remove_pid, SIGTERM, NULL);
 		ret = -E_SIGNAL;
 		break;
 	case SIGHUP:
@@ -968,22 +1025,48 @@ static int use_rsync_locally(char *logname)
 static int rename_resume_snap(int64_t creation_time)
 {
 	struct snapshot_list sl = {.num_snapshots = 0};
-	struct snapshot *s;
+	struct snapshot *s = NULL;
 	char *new_name = incomplete_name(creation_time);
 	int ret;
+	const char *why;
 
 	ret = 0;
 	if (conf.no_resume_given)
 		goto out;
 	dss_get_snapshot_list(&sl);
+	/*
+	 * Snapshot recycling: We first look at the newest snapshot. If this
+	 * snapshot happens to be incomplete, the last rsync process was
+	 * aborted and we reuse this one. Otherwise we look at snapshots which
+	 * could be removed (outdated and redundant snapshots) as candidates
+	 * for recycling. If no outdated/redundant snapshot exists, we check if
+	 * there is an orphaned snapshot, which likely is useless anyway.
+	 *
+	 * Only if no existing snapshot is suitable for recycling, we bite the
+	 * bullet and create a new one.
+	 */
 	s = get_newest_snapshot(&sl);
-	if (!s)
+	if (!s) /* no snapshots at all */
 		goto out;
-	if ((s->flags & SS_COMPLETE) != 0) /* complete */
+	/* re-use last snapshot if it is incomplete */
+	why = "aborted";
+	if ((s->flags & SS_COMPLETE) == 0)
 		goto out;
-	DSS_INFO_LOG("resuming: reusing %s as destination dir\n", s->name);
-	ret = dss_rename(s->name, new_name);
+	why = "outdated";
+	s = find_outdated_snapshot(&sl);
+	if (s)
+		goto out;
+	why = "redundant";
+	s = find_redundant_snapshot(&sl);
+	if (s)
+		goto out;
+	why = "orphaned";
+	s = find_orphaned_snapshot(&sl);
 out:
+	if (s) {
+		DSS_INFO_LOG("reusing %s snapshot %s\n", why, s->name);
+		ret = dss_rename(s->name, new_name);
+	}
 	if (ret >= 0)
 		DSS_NOTICE_LOG("creating new snapshot %s\n", new_name);
 	free(new_name);
@@ -1042,14 +1125,12 @@ static void free_rsync_argv(char **argv)
 
 static int create_snapshot(char **argv)
 {
-	int ret, fds[3] = {0, 0, 0};
+	int ret;
 
 	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;
+	dss_exec(&create_pid, argv[0], argv);
 	snapshot_creation_status = HS_RUNNING;
 	return ret;
 }
@@ -1074,7 +1155,6 @@ static int select_loop(void)
 		}
 		FD_ZERO(&rfds);
 		FD_SET(signal_pipe, &rfds);
-		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)
 			goto out;
@@ -1092,9 +1172,7 @@ static int select_loop(void)
 			continue;
 		}
 		if (snapshot_removal_status == HS_SUCCESS) {
-			ret = post_remove_hook();
-			if (ret < 0)
-				goto out;
+			post_remove_hook();
 			continue;
 		}
 		ret = try_to_free_disk_space();
@@ -1109,9 +1187,7 @@ static int select_loop(void)
 		case HS_READY:
 			if (!next_snapshot_is_due())
 				continue;
-			ret = pre_create_hook();
-			if (ret < 0)
-				goto out;
+			pre_create_hook();
 			continue;
 		case HS_PRE_RUNNING:
 		case HS_RUNNING:
@@ -1122,7 +1198,10 @@ static int select_loop(void)
 				free_rsync_argv(rsync_argv);
 				create_rsync_argv(&rsync_argv, &current_snapshot_creation_time);
 			}
-			/* fall through */
+			ret = create_snapshot(rsync_argv);
+			if (ret < 0)
+				goto out;
+			continue;
 		case HS_NEEDS_RESTART:
 			if (!next_snapshot_is_due())
 				continue;
@@ -1131,9 +1210,7 @@ static int select_loop(void)
 				goto out;
 			continue;
 		case HS_SUCCESS:
-			ret = post_create_hook();
-			if (ret < 0)
-				goto out;
+			post_create_hook();
 			continue;
 		}
 	}
@@ -1143,14 +1220,15 @@ out:
 
 static void exit_hook(int exit_code)
 {
-	int fds[3] = {0, 0, 0};
-	char *argv[] = {conf.exit_hook_arg, dss_strerror(-exit_code), NULL};
+	char *argv[3];
 	pid_t pid;
 
-	if (!conf.exit_hook_given)
-		return;
+	argv[0] = conf.exit_hook_arg;
+	argv[1] = dss_strerror(-exit_code);
+	argv[2] = NULL;
+
 	DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
-	dss_exec(&pid, conf.exit_hook_arg, argv, fds);
+	dss_exec(&pid, conf.exit_hook_arg, argv);
 }
 
 static int com_run(void)
@@ -1201,9 +1279,7 @@ rm:
 		ret = 0;
 		goto out;
 	}
-	ret = pre_remove_hook(victim, why);
-	if (ret < 0)
-		goto out;
+	pre_remove_hook(victim, why);
 	if (snapshot_removal_status == HS_PRE_RUNNING) {
 		ret = wait_for_remove_process();
 		if (ret < 0)
@@ -1219,9 +1295,7 @@ rm:
 		goto out;
 	if (snapshot_removal_status != HS_SUCCESS)
 		goto out;
-	ret = post_remove_hook();
-	if (ret < 0)
-		goto out;
+	post_remove_hook();
 	if (snapshot_removal_status != HS_POST_RUNNING)
 		goto out;
 	ret = wait_for_remove_process();
@@ -1253,9 +1327,7 @@ static int com_create(void)
 		free(msg);
 		return 1;
 	}
-	ret = pre_create_hook();
-	if (ret < 0)
-		return ret;
+	pre_create_hook();
 	if (create_pid) {
 		ret = wait_for_process(create_pid, &status);
 		if (ret < 0)
@@ -1351,10 +1423,10 @@ int main(int argc, char **argv)
 	}
 	if (conf.daemon_given)
 		daemon_init();
-	dump_dss_config("dss configuration");
 	ret = change_to_dest_dir();
 	if (ret < 0)
 		goto out;
+	dump_dss_config("startup");
 	ret = setup_signal_handling();
 	if (ret < 0)
 		goto out;