X-Git-Url: http://git.tuebingen.mpg.de/?p=dss.git;a=blobdiff_plain;f=dss.c;h=217dfafd25c55741e564c0feb13c1ccfbad317d0;hp=24d77eb79dbd2ab0862b2b7012b73babde542c93;hb=HEAD;hpb=07ab37b55e1bf4ad7ef65fd25db84fa2a5f50f56

diff --git a/dss.c b/dss.c
index 24d77eb..0992ec6 100644
--- a/dss.c
+++ b/dss.c
@@ -1,8 +1,4 @@
-/*
- * Copyright (C) 2008-2011 Andre Noll <maan@tuebingen.mpg.de>
- *
- * Licensed under the GPL v2. For licencing details see COPYING.
- */
+/* SPDX-License-Identifier: GPL-2.0 */
 #include <string.h>
 #include <stdlib.h>
 #include <stdio.h>
@@ -85,7 +81,7 @@ static int64_t next_snapshot_time;
 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. */
+/* Set by the pre-rm hook, cleared by handle_remove_exit(). */
 struct snapshot *snapshot_currently_being_removed;
 /** Needed by the post-create hook. */
 static char *path_to_last_complete_snapshot;
@@ -162,7 +158,7 @@ static void dump_dss_config(const char *msg)
 	fprintf(log, "\n*** internal state ***\n\n");
 	fprintf(log,
 		"pid: %d\n"
-		"logile: %s\n"
+		"logfile: %s\n"
 		"snapshot_currently_being_removed: %s\n"
 		"path_to_last_complete_snapshot: %s\n"
 		"reference_snapshot: %s\n"
@@ -287,11 +283,13 @@ static char *get_config_file_name(void)
 	return config_file;
 }
 
-static int send_signal(int sig)
+static int send_signal(int sig, bool wait)
 {
 	pid_t pid;
 	char *config_file = get_config_file_name();
 	int ret = get_dss_pid(config_file, &pid);
+	unsigned ms = 32;
+	struct timespec ts;
 
 	free(config_file);
 	if (ret < 0)
@@ -304,7 +302,23 @@ static int send_signal(int sig)
 	ret = kill(pid, sig);
 	if (ret < 0)
 		return -ERRNO_TO_DSS_ERROR(errno);
-	return 1;
+	if (!wait)
+		return 1;
+	while (ms < 5000) {
+		ts.tv_sec = ms / 1000;
+		ts.tv_nsec = (ms % 1000) * 1000 * 1000;
+		ret = nanosleep(&ts, NULL);
+		if (ret < 0)
+			return -ERRNO_TO_DSS_ERROR(errno);
+		ret = kill(pid, 0);
+		if (ret < 0) {
+			if (errno != ESRCH)
+				return -ERRNO_TO_DSS_ERROR(errno);
+			return 1;
+		}
+		ms *= 2;
+	}
+	return -E_KILL_TIMEOUT;
 }
 
 struct signal_info {
@@ -363,6 +377,7 @@ static const struct signal_info signal_table[] = {
 
 static int com_kill(void)
 {
+	bool w_given = OPT_GIVEN(KILL, WAIT);
 	const char *arg = OPT_STRING_VAL(KILL, SIGNAL);
 	int ret, i;
 
@@ -373,17 +388,17 @@ static int com_kill(void)
 			return ret;
 		if (val < 0 || val > SIGRTMAX)
 			return -ERRNO_TO_DSS_ERROR(EINVAL);
-		return send_signal(val);
+		return send_signal(val, w_given);
 	}
 	if (strncasecmp(arg, "sig", 3) == 0)
 		arg += 3;
 	if (strcasecmp(arg, "CLD") == 0)
-		return send_signal(SIGCHLD);
+		return send_signal(SIGCHLD, w_given);
 	if (strcasecmp(arg, "IOT") == 0)
-		return send_signal(SIGABRT);
+		return send_signal(SIGABRT, w_given);
 	for (i = 0; i < SIGNAL_TABLE_SIZE; i++)
 		if (strcasecmp(arg, signal_table[i].name) == 0)
-			return send_signal(signal_table[i].num);
+			return send_signal(signal_table[i].num, w_given);
 	DSS_ERROR_LOG(("invalid sigspec: %s\n", arg));
 	return -ERRNO_TO_DSS_ERROR(EINVAL);
 }
@@ -398,7 +413,8 @@ static void dss_get_snapshot_list(struct snapshot_list *sl)
 static int64_t compute_next_snapshot_time(void)
 {
 	int64_t x = 0, now = get_current_time(), unit_interval
-		= 24 * 3600 * OPT_UINT32_VAL(DSS, UNIT_INTERVAL), ret;
+		= 24 * 3600 * OPT_UINT32_VAL(DSS, UNIT_INTERVAL), ret,
+		last_completion_time;
 	unsigned wanted = desired_number_of_snapshots(0,
 		OPT_UINT32_VAL(DSS, NUM_INTERVALS)),
 		num_complete = 0;
@@ -412,6 +428,7 @@ static int64_t compute_next_snapshot_time(void)
 			continue;
 		num_complete++;
 		x += s->completion_time - s->creation_time;
+		last_completion_time = s->completion_time;
 	}
 	assert(x >= 0);
 
@@ -421,7 +438,7 @@ static int64_t compute_next_snapshot_time(void)
 	x /= num_complete; /* 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;
+	ret = last_completion_time + unit_interval / wanted - x;
 out:
 	free_snapshot_list(&sl);
 	return ret;
@@ -557,9 +574,6 @@ static int is_reference_snapshot(struct snapshot *s)
 	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 i, interval;
@@ -637,6 +651,7 @@ static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 	int i, num_complete;
 	struct snapshot *s, *ref = NULL;
 
+	DSS_DEBUG_LOG(("picking snapshot with earliest creation time\n"));
 	num_complete = num_complete_snapshots(sl);
 	if (num_complete <= OPT_UINT32_VAL(DSS, MIN_COMPLETE))
 		return NULL;
@@ -647,7 +662,6 @@ static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 			ref = s;
 			continue;
 		}
-		DSS_INFO_LOG(("oldest removable snapshot: %s\n", s->name));
 		return s;
 	}
 	assert(ref);
@@ -655,6 +669,50 @@ static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 	return ref;
 }
 
+/* returns NULL <==> *reason is set to NULL */
+static struct snapshot *find_removable_snapshot(struct snapshot_list *sl,
+		bool try_hard, char **reason)
+{
+	struct snapshot *victim;
+
+	/*
+	 * Don't remove anything if there is free space and we have fewer
+	 * snapshots than configured, plus one. This way there is always one
+	 * snapshot that can be recycled.
+	 */
+	if (!try_hard && sl->num_snapshots <=
+			1 << OPT_UINT32_VAL(DSS, NUM_INTERVALS))
+		goto nope;
+	victim = find_orphaned_snapshot(sl);
+	if (victim) {
+		*reason = make_message("orphaned");
+		return victim;
+	}
+	victim = find_outdated_snapshot(sl);
+	if (victim) {
+		*reason = make_message("outdated");
+		return victim;
+	}
+	if (!OPT_GIVEN(DSS, KEEP_REDUNDANT)) {
+		victim = find_redundant_snapshot(sl);
+		if (victim) {
+			*reason = make_message("redundant");
+			return victim;
+		}
+	}
+	if (!try_hard)
+		goto nope;
+	DSS_WARNING_LOG(("nothing obvious to remove\n"));
+	victim = find_oldest_removable_snapshot(sl);
+	if (victim) {
+		*reason = make_message("oldest");
+		return victim;
+	}
+nope:
+	*reason = NULL;
+	return NULL;
+}
+
 static int rename_incomplete_snapshot(int64_t start)
 {
 	char *old_name;
@@ -687,7 +745,7 @@ static int try_to_free_disk_space(void)
 	struct snapshot_list sl;
 	struct snapshot *victim;
 	struct timeval now;
-	const char *why;
+	char *why;
 	int low_disk_space;
 
 	ret = disk_space_low(NULL);
@@ -698,55 +756,25 @@ static int try_to_free_disk_space(void)
 	if (tv_diff(&next_removal_check, &now, NULL) > 0)
 		return 0;
 	if (!low_disk_space) {
-		if (OPT_GIVEN(DSS, KEEP_REDUNDANT))
-			return 0;
 		if (snapshot_creation_status != HS_READY)
 			return 0;
 		if (next_snapshot_is_due())
 			return 0;
 	}
-	/*
-	 * Idle and --keep_redundant not given, or low disk space. Look at
-	 * existing snapshots.
-	 */
+	/* Idle or low disk space, look at existing snapshots. */
 	dss_get_snapshot_list(&sl);
-	ret = 0;
-	/*
-	 * Don't remove anything if there is free space and we have fewer
-	 * snapshots than configured, plus one. This way there is always one
-	 * snapshot that can be recycled.
-	 */
-	if (!low_disk_space && sl.num_snapshots <=
-			1 << OPT_UINT32_VAL(DSS, NUM_INTERVALS))
-		goto out;
-	why = "outdated";
-	victim = find_outdated_snapshot(&sl);
-	if (victim)
-		goto remove;
-	why = "redundant";
-	victim = find_redundant_snapshot(&sl);
-	if (victim)
-		goto remove;
-	why = "orphaned";
-	victim = find_orphaned_snapshot(&sl);
+	victim = find_removable_snapshot(&sl, low_disk_space, &why);
+	if (victim) {
+		pre_remove_hook(victim, why);
+		free(why);
+	}
+	free_snapshot_list(&sl);
 	if (victim)
-		goto remove;
-	/* try harder only if disk space is low */
+		return 1;
 	if (!low_disk_space)
-		goto out;
-	DSS_WARNING_LOG(("disk space low and nothing obvious to remove\n"));
-	why = "oldest";
-	victim = find_oldest_removable_snapshot(&sl);
-	if (victim)
-		goto remove;
+		return 0;
 	DSS_CRIT_LOG(("uhuhu: disk space low and nothing to remove\n"));
-	ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
-	goto out;
-remove:
-	pre_remove_hook(victim, why);
-out:
-	free_snapshot_list(&sl);
-	return ret;
+	return -ERRNO_TO_DSS_ERROR(ENOSPC);
 }
 
 static void post_create_hook(void)
@@ -1094,7 +1122,6 @@ static int change_to_dest_dir(void)
 
 static int check_config(void)
 {
-	int ret;
 	uint32_t unit_interval = OPT_UINT32_VAL(DSS, UNIT_INTERVAL);
 	uint32_t num_intervals = OPT_UINT32_VAL(DSS, NUM_INTERVALS);
 
@@ -1119,9 +1146,6 @@ static int check_config(void)
 			DSS_ERROR_LOG(("--dest-dir required\n"));
 			return -E_SYNTAX;
 		}
-		ret = change_to_dest_dir();
-		if (ret < 0)
-			return ret;
 	}
 	DSS_DEBUG_LOG(("number of intervals: %i\n", num_intervals));
 	return 1;
@@ -1235,8 +1259,6 @@ close_fd:
 		close(fd);
 out:
 	free(config_file);
-	if (ret < 0)
-		DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
 	return ret;
 }
 
@@ -1284,9 +1306,7 @@ static int handle_signal(void)
 	switch (sig) {
 	case SIGINT:
 	case SIGTERM:
-		kill_children();
-		ret = -E_SIGNAL;
-		break;
+		return -E_SIGNAL;
 	case SIGHUP:
 		ret = handle_sighup();
 		break;
@@ -1371,7 +1391,7 @@ out:
 static void create_rsync_argv(char ***argv, int64_t *num)
 {
 	char *logname;
-	int i = 0, j, N = OPT_GIVEN(DSS, RSYNC_OPTION);
+	int i = 0, j, N;
 	struct snapshot_list sl;
 	static bool seeded;
 
@@ -1380,7 +1400,13 @@ static void create_rsync_argv(char ***argv, int64_t *num)
 	name_of_reference_snapshot = name_of_newest_complete_snapshot(&sl);
 	free_snapshot_list(&sl);
 
-	*argv = dss_malloc((15 + N) * sizeof(char *));
+	/*
+	 * We specify up to 6 arguments, one argument per given rsync option
+	 * and one argument per given source dir. We also need space for the
+	 * terminating NULL pointer.
+	 */
+	N = OPT_GIVEN(DSS, RSYNC_OPTION) + OPT_GIVEN(DSS, SOURCE_DIR);
+	*argv = dss_malloc((7 + N) * sizeof(char *));
 	(*argv)[i++] = dss_strdup("rsync");
 	(*argv)[i++] = dss_strdup("-a");
 	(*argv)[i++] = dss_strdup("--delete");
@@ -1392,7 +1418,7 @@ static void create_rsync_argv(char ***argv, int64_t *num)
 		DSS_NOTICE_LOG(("adding --checksum to rsync options\n"));
 		(*argv)[i++] = dss_strdup("--checksum");
 	}
-	for (j = 0; j < N; j++)
+	for (j = 0; j < OPT_GIVEN(DSS, RSYNC_OPTION); j++)
 		(*argv)[i++] = dss_strdup(lls_string_val(j,
 			OPT_RESULT(DSS, RSYNC_OPTION)));
 	if (name_of_reference_snapshot) {
@@ -1402,14 +1428,34 @@ static void create_rsync_argv(char ***argv, int64_t *num)
 	} else
 		DSS_INFO_LOG(("no suitable reference snapshot found\n"));
 	logname = dss_logname();
-	if (use_rsync_locally(logname))
-		(*argv)[i++] = dss_strdup(OPT_STRING_VAL(DSS, SOURCE_DIR));
-	else
-		(*argv)[i++] = make_message("%s@%s:%s/",
-			OPT_GIVEN(DSS, REMOTE_USER)?
-				OPT_STRING_VAL(DSS, REMOTE_USER) : logname,
-			OPT_STRING_VAL(DSS, REMOTE_HOST),
-			OPT_STRING_VAL(DSS, SOURCE_DIR));
+	if (use_rsync_locally(logname)) {
+		for (j = 0; j < OPT_GIVEN(DSS, SOURCE_DIR); j++)
+			(*argv)[i++] = dss_strdup(lls_string_val(j,
+				OPT_RESULT(DSS, SOURCE_DIR)));
+	} else {
+		/*
+		 * dss-1.0 and earlier did not support multiple source
+		 * directories.  These versions appended a slash to the end of
+		 * the source directory to make sure that only the contents of
+		 * the single source directory, but not the directory itself,
+		 * are copied to the destination. For multiple source
+		 * directories, however, this is not a good idea because the
+		 * source directories may well contain identical file names,
+		 * which would then be copied to the same location on the
+		 * destination, overwriting each other. Moreover, we want the
+		 * directory on the destination match the source. To preserve
+		 * the old behaviour, we thus have to special-case N=1.
+		 */
+		for (j = 0; j < OPT_GIVEN(DSS, SOURCE_DIR); j++) {
+			(*argv)[i++] = make_message("%s@%s:%s%s",
+				OPT_GIVEN(DSS, REMOTE_USER)?
+					OPT_STRING_VAL(DSS, REMOTE_USER) : logname,
+				OPT_STRING_VAL(DSS, REMOTE_HOST),
+				lls_string_val(j, OPT_RESULT(DSS, SOURCE_DIR)),
+				OPT_GIVEN(DSS, SOURCE_DIR) == 1? "/" : ""
+			);
+		}
+	}
 	free(logname);
 	*num = get_current_time();
 	(*argv)[i++] = incomplete_name(*num);
@@ -1433,6 +1479,7 @@ static int create_snapshot(char **argv)
 {
 	int ret;
 
+	assert(argv);
 	ret = rename_resume_snap(current_snapshot_creation_time);
 	if (ret < 0)
 		return ret;
@@ -1526,15 +1573,18 @@ out:
 
 static void exit_hook(int exit_code)
 {
-	const char *argv[3];
 	pid_t pid;
-
-	argv[0] = OPT_STRING_VAL(DSS, EXIT_HOOK);
-	argv[1] = dss_strerror(-exit_code);
-	argv[2] = NULL;
-
-	DSS_NOTICE_LOG(("executing %s %s\n", argv[0], argv[1]));
-	dss_exec(&pid, argv[0], (char **)argv);
+	char **argv, *tmp = dss_strdup(OPT_STRING_VAL(DSS, EXIT_HOOK));
+	unsigned n = split_args(tmp, &argv);
+
+	n++;
+	argv = dss_realloc(argv, (n + 1) * sizeof(char *));
+	argv[n - 1] = dss_strdup(dss_strerror(-exit_code));
+	argv[n] = NULL;
+	dss_exec(&pid, argv[0], argv);
+	free(argv[n - 1]);
+	free(argv);
+	free(tmp);
 }
 
 static void lock_dss_or_die(void)
@@ -1566,12 +1616,23 @@ static int com_run(void)
 		DSS_ERROR_LOG(("pid %d\n", (int)pid));
 		return -E_ALREADY_RUNNING;
 	}
+	/*
+	 * Order is important here: Since daemon_init() forks, it would drop
+	 * the lock if it had been acquired already. Changing the cwd before
+	 * grabbing the lock causes stat(2) to fail in case a relative config
+	 * file path was given, which results in a different key ID for
+	 * locking. Therefore we must first daemonize, then lock, then change
+	 * the cwd.
+	 */
 	if (OPT_GIVEN(RUN, DAEMON)) {
 		fd = daemon_init();
 		daemonized = true;
 		logfile = open_log(OPT_STRING_VAL(RUN, LOGFILE));
 	}
 	lock_dss_or_die();
+	ret = change_to_dest_dir();
+	if (ret < 0)
+		return ret;
 	dump_dss_config("startup");
 	ret = install_sighandler(SIGHUP);
 	if (ret < 0)
@@ -1591,6 +1652,8 @@ static int com_run(void)
 		ret = -E_BUG;
 	kill_children();
 	exit_hook(ret);
+	while (wait(NULL) >= 0 || errno != ECHILD)
+		; /* still have children to wait for */
 	return ret;
 }
 EXPORT_CMD_HANDLER(run);
@@ -1601,55 +1664,58 @@ static int com_prune(void)
 	struct snapshot_list sl;
 	struct snapshot *victim;
 	struct disk_space ds;
-	const char *why;
+	char *why;
+	bool try_hard;
 
 	lock_dss_or_die();
-	ret = get_disk_space(".", &ds);
+	ret = change_to_dest_dir();
 	if (ret < 0)
 		return ret;
-	log_disk_space(&ds);
+	switch (OPT_UINT32_VAL(PRUNE, DISK_SPACE)) {
+	case FDS_LOW: try_hard = true; break;
+	case FDS_HIGH: try_hard = false; break;
+	default:
+		ret = get_disk_space(".", &ds);
+		if (ret < 0)
+			return ret;
+		log_disk_space(&ds);
+		try_hard = disk_space_low(&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:
+	victim = find_removable_snapshot(&sl, try_hard, &why);
+	if (!victim) {
+		dss_msg("nothing to prune\n");
+		ret = 0;
+		goto free_sl;
+	}
 	if (OPT_GIVEN(DSS, DRY_RUN)) {
-		dss_msg("%s snapshot %s (interval = %i)\n",
+		dss_msg("picking %s snapshot %s (interval = %i)\n",
 			why, victim->name, victim->interval);
 		ret = 0;
-		goto out;
+		goto free_why;
 	}
 	pre_remove_hook(victim, why);
 	if (snapshot_removal_status == HS_PRE_RUNNING) {
 		ret = wait_for_remove_process();
 		if (ret < 0)
-			goto out;
+			goto free_why;
+		ret = -E_HOOK_FAILED;
 		if (snapshot_removal_status != HS_PRE_SUCCESS)
-			goto out;
+			goto free_why;
 	}
 	ret = exec_rm();
 	if (ret < 0)
-		goto out;
+		goto free_why;
 	ret = wait_for_remove_process();
 	if (ret < 0)
-		goto out;
-	if (snapshot_removal_status != HS_SUCCESS)
-		goto out;
+		goto free_why;
+	assert(snapshot_removal_status == HS_SUCCESS);
 	post_remove_hook();
-	if (snapshot_removal_status != HS_POST_RUNNING)
-		goto out;
+	assert(snapshot_removal_status == HS_POST_RUNNING);
 	ret = wait_for_remove_process();
-	if (ret < 0)
-		goto out;
-	ret = 1;
-out:
+free_why:
+	free(why);
+free_sl:
 	free_snapshot_list(&sl);
 	return ret;
 }
@@ -1661,6 +1727,9 @@ static int com_create(void)
 	char **rsync_argv;
 
 	lock_dss_or_die();
+	ret = change_to_dest_dir();
+	if (ret < 0)
+		return ret;
 	if (OPT_GIVEN(DSS, DRY_RUN)) {
 		int i;
 		char *msg = NULL;
@@ -1706,11 +1775,14 @@ EXPORT_CMD_HANDLER(create);
 
 static int com_ls(void)
 {
-	int i;
+	int i, ret;
 	struct snapshot_list sl;
 	struct snapshot *s;
 	int64_t now = get_current_time();
 
+	ret = change_to_dest_dir();
+	if (ret < 0)
+		return ret;
 	dss_get_snapshot_list(&sl);
 	FOR_EACH_SNAPSHOT(s, i, &sl) {
 		int64_t d;
@@ -1757,7 +1829,7 @@ static void handle_version_and_help(void)
 	else if (OPT_GIVEN(DSS, HELP))
 		txt = lls_short_help(CMD_PTR(DSS));
 	else if (OPT_GIVEN(DSS, VERSION))
-		txt = dss_strdup(VERSION_STRING);
+		txt = make_message("%s\n", VERSION_STRING);
 	else
 		return;
 	printf("%s", txt);