Improve error diagnostics.
[dss.git] / dss.c
diff --git a/dss.c b/dss.c
index 3bcf60d4ba39d083f27bd26cb85998bb7a6eff69..a233c11f7060c0321502c35e555df78ef0b3931b 100644 (file)
--- a/dss.c
+++ b/dss.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (C) 2008 Andre Noll <maan@systemlinux.org>
+ * Copyright (C) 2008-2009 Andre Noll <maan@systemlinux.org>
  *
  * Licensed under the GPL v2. For licencing details see COPYING.
  */
@@ -47,15 +47,16 @@ static int create_process_stopped;
 /** Process id of current pre-remove/rm/post-remove process. */
 static pid_t remove_pid;
 /** When the next snapshot is due. */
-static struct timeval next_snapshot_time;
+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. */
+/** 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. */
@@ -150,21 +151,16 @@ static void dss_get_snapshot_list(struct snapshot_list *sl)
        get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
 }
 
-static int next_snapshot_is_due(void)
+static int64_t compute_next_snapshot_time(void)
 {
-       struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
-               tmp, diff;
-       int64_t x = 0;
+       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, ret;
+       int i;
        struct snapshot *s = NULL;
        struct snapshot_list sl;
 
-       gettimeofday(&now, NULL);
-       if (tv_diff(&next_snapshot_time, &now, NULL) > 0)
-               return 0;
-       assert(snapshot_creation_status == HS_READY);
        current_snapshot_creation_time = 0;
        dss_get_snapshot_list(&sl);
        FOR_EACH_SNAPSHOT(s, i, &sl) {
@@ -174,37 +170,52 @@ static int next_snapshot_is_due(void)
                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 */
-       if (ret < 0 || !s) { /* unit_interval < tmp or no snapshot */
-               ret = 1;
-               goto out;
-       }
 
-       tv_divide(wanted, &diff, &tmp); /* sleep time between two snapshots */
-       diff.tv_sec = s->completion_time; /* completion time of the latest snapshot */
-       diff.tv_usec = 0;
-       tv_add(&diff, &tmp, &next_snapshot_time);
-       ret = (tv_diff(&now, &next_snapshot_time, &diff) < 0)? 0 : 1;
+       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);
-       if (ret == 0)
-               DSS_DEBUG_LOG("next snapshot due in %lu seconds\n",
-                       tv2ms(&diff) / 1000);
-       else
-               DSS_DEBUG_LOG("next snapshot: now\n");
        return ret;
 }
 
+static inline void invalidate_next_snapshot_time(void)
+{
+       next_snapshot_time = 0;
+}
+
+static inline int next_snapshot_time_is_valid(void)
+{
+       return next_snapshot_time != 0;
+}
+
+static int next_snapshot_is_due(void)
+{
+       int64_t now = get_current_time();
+
+       assert(snapshot_creation_status == HS_READY);
+       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;
+       }
+       DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n",
+               next_snapshot_time - now);
+       return 0;
+}
+
 static int 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;
@@ -218,11 +229,13 @@ static int pre_create_hook(void)
        return ret;
 }
 
-static int pre_remove_hook(struct snapshot *s, char *why)
+static int pre_remove_hook(struct snapshot *s, const char *why)
 {
        int ret, fds[3] = {0, 0, 0};
        char *cmd;
 
+       if (!s)
+               return 0;
        DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
        assert(snapshot_removal_status == HS_READY);
        assert(remove_pid == 0);
@@ -277,12 +290,53 @@ 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 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;
 
@@ -305,6 +359,8 @@ static int remove_redundant_snapshot(struct snapshot_list *sl)
 
                        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;
@@ -329,56 +385,43 @@ static int remove_redundant_snapshot(struct snapshot_list *sl)
                        prev = s;
                }
                assert(victim);
-               if (conf.dry_run_given) {
-                       dss_msg("%s would be removed (interval = %i)\n",
-                               victim->name, victim->interval);
-                       continue;
-               }
-               ret = pre_remove_hook(victim, "redundant");
-               return ret < 0? ret : 1;
+               return victim;
        }
-       return 0;
+       return NULL;
 }
 
-static int remove_outdated_snapshot(struct snapshot_list *sl)
+static struct snapshot *find_outdated_snapshot(struct snapshot_list *sl)
 {
-       int i, ret;
+       int i;
        struct snapshot *s;
 
-       DSS_DEBUG_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 (snapshot_is_being_created(s))
                        continue;
-               if (s->interval <= conf.num_intervals_arg)
+               if (is_reference_snapshot(s))
                        continue;
-               if (conf.dry_run_given) {
-                       dss_msg("%s would be removed (interval = %i)\n",
-                               s->name, s->interval);
+               if (s->interval < conf.num_intervals_arg)
                        continue;
-               }
-               ret = pre_remove_hook(s, "outdated");
-               if (ret < 0)
-                       return ret;
-               return 1;
+               return s;
        }
-       return 0;
+       return NULL;
 }
 
-static int remove_oldest_snapshot(struct snapshot_list *sl)
+struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
 {
-       int ret;
-       struct snapshot *s = get_oldest_snapshot(sl);
-
-       if (!s) /* no snapshot found */
-               return 0;
-       DSS_INFO_LOG("oldest snapshot: %s\n", s->name);
-       if (snapshot_is_being_created(s))
-               return 0;
-       ret = pre_remove_hook(s, "oldest");
-       if (ret < 0)
-               return ret;
-       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;
 }
 
 static int rename_incomplete_snapshot(int64_t start)
@@ -400,34 +443,58 @@ static int rename_incomplete_snapshot(int64_t start)
        return ret;
 }
 
-static int try_to_free_disk_space(int low_disk_space)
+static int try_to_free_disk_space(void)
 {
        int ret;
        struct snapshot_list sl;
+       struct snapshot *victim;
        struct timeval now;
+       const char *why;
+       int low_disk_space;
 
+       ret = disk_space_low();
+       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 && conf.keep_redundant_given)
-               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;
+       }
        dss_get_snapshot_list(&sl);
-       ret = remove_outdated_snapshot(&sl);
-       if (ret) /* error, or we are removing something */
-               goto out;
-       /* no outdated snapshot */
-       ret = remove_redundant_snapshot(&sl);
-       if (ret)
-               goto out;
        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");
-       ret = remove_oldest_snapshot(&sl);
-       if (ret)
-               goto out;
-       DSS_CRIT_LOG("uhuhu: not enough disk space for a single snapshot\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;
@@ -650,8 +717,7 @@ static int handle_rsync_exit(int status)
                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;
+               next_snapshot_time = get_current_time() + 60;
                ret = 1;
                goto out;
        }
@@ -665,6 +731,8 @@ static int handle_rsync_exit(int status)
        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;
@@ -689,6 +757,7 @@ static int handle_pre_create_hook_exit(int status)
                        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;
@@ -749,9 +818,13 @@ static int check_config(void)
        return 1;
 }
 
+/*
+ * 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;
@@ -770,13 +843,13 @@ static int parse_config_file(int override)
                old_daemon_given = conf.daemon_given;
        }
 
-       ret = stat(config_file, &statbuf);
-       if (ret && conf.config_file_given) {
+       config_file_exists = !stat(config_file, &statbuf);
+       if (!config_file_exists && conf.config_file_given) {
                ret = -ERRNO_TO_DSS_ERROR(errno);
                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 = override,
                        .initialize = 0,
@@ -816,6 +889,7 @@ static int parse_config_file(int override)
        }
        DSS_DEBUG_LOG("loglevel: %d\n", conf.loglevel_arg);
 //     cmdline_parser_dump(logfile? logfile : stdout, &conf);
+       ret = config_file_exists;
 out:
        free(config_file);
        if (ret < 0)
@@ -837,6 +911,7 @@ static int handle_sighup(void)
        ret = parse_config_file(1);
        if (ret < 0)
                return ret;
+       invalidate_next_snapshot_time();
        return change_to_dest_dir();
 }
 
@@ -885,12 +960,13 @@ static int use_rsync_locally(char *logname)
 
 static void create_rsync_argv(char ***argv, int64_t *num)
 {
-       char *logname, *newest;
+       char *logname;
        int i = 0, j;
        struct snapshot_list sl;
 
        dss_get_snapshot_list(&sl);
-       newest = name_of_newest_complete_snapshot(&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 *));
@@ -899,12 +975,12 @@ static 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");
+               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);
@@ -955,7 +1031,6 @@ static int select_loop(void)
 
        for (;;) {
                fd_set rfds;
-               int low_disk_space;
                struct timeval *tvp;
 
                if (remove_pid)
@@ -990,11 +1065,7 @@ static int select_loop(void)
                                goto out;
                        continue;
                }
-               ret = disk_space_low();
-               if (ret < 0)
-                       goto out;
-               low_disk_space = ret;
-               ret = try_to_free_disk_space(low_disk_space);
+               ret = try_to_free_disk_space();
                if (ret < 0)
                        goto out;
                if (snapshot_removal_status != HS_READY) {
@@ -1068,26 +1139,35 @@ 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);
-       ret = remove_outdated_snapshot(&sl);
-       if (ret < 0)
-               goto out;
-       if (ret > 0)
+       why = "outdated";
+       victim = find_outdated_snapshot(&sl);
+       if (victim)
                goto rm;
-       ret = remove_redundant_snapshot(&sl);
-       if (ret < 0)
-               goto out;
-       if (ret > 0)
+       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 = 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)
@@ -1219,6 +1299,20 @@ int main(int argc, char **argv)
        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 = change_to_dest_dir();