/*
- * Copyright (C) 2008 Andre Noll <maan@systemlinux.org>
+ * Copyright (C) 2008-2011 Andre Noll <maan@systemlinux.org>
*
* Licensed under the GPL v2. For licencing details see COPYING.
*/
static FILE *logfile;
/** The read end of the signal pipe */
static int signal_pipe;
-/** Process id of current rsync process. */
-static pid_t rsync_pid;
-/** Whether the rsync process is currently stopped */
-static int rsync_stopped;
-/** Process id of current rm process. */
-static pid_t rm_pid;
+/** Process id of current pre-create-hook/rsync/post-create-hook process. */
+static pid_t create_pid;
+/** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
+static int create_process_stopped;
+/** Process id of current pre-remove/rm/post-remove process. */
+static pid_t remove_pid;
/** When the next snapshot is due. */
-static struct timeval next_snapshot_time;
-/** The pid of the pre-create hook. */
-static pid_t pre_create_hook_pid;
-/** The pid of the post-create hook. */
-static pid_t post_create_hook_pid;
+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. */
-static unsigned snapshot_creation_status;
+enum hook_status snapshot_creation_status;
+/** \sa \ref snap.h for details. */
+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 \
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);
}
-static void compute_next_snapshot_time(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;
- assert(snapshot_creation_status == SCS_READY);
- current_snapshot_creation_time = 0;
dss_get_snapshot_list(&sl);
FOR_EACH_SNAPSHOT(s, i, &sl) {
if (!(s->flags & SS_COMPLETE))
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 */
- goto min_sleep;
- tv_divide(wanted, &diff, &tmp); /* sleep time betweeen two snapshots */
- diff.tv_sec = s->completion_time; /* completion time of the the latest snaphot */
- diff.tv_usec = 0;
- tv_add(&diff, &tmp, &next_snapshot_time);
- gettimeofday(&now, NULL);
- if (tv_diff(&now, &next_snapshot_time, NULL) < 0)
+
+ 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;
-min_sleep:
- next_snapshot_time = now;
- next_snapshot_time.tv_sec += 60;
+ ret = s->completion_time + unit_interval / wanted - x;
out:
free_snapshot_list(&sl);
+ 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();
+
+ 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 void pre_create_hook(void)
+{
+ 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);
+ dss_exec_cmdline_pid(&create_pid, conf.pre_create_hook_arg);
+ snapshot_creation_status = HS_PRE_RUNNING;
+}
+
+static void pre_remove_hook(struct snapshot *s, const char *why)
+{
+ char *cmd;
+
+ if (!s)
+ return;
+ 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);
+ dss_exec_cmdline_pid(&remove_pid, cmd);
+ free(cmd);
+ snapshot_removal_status = HS_PRE_RUNNING;
+}
-static int remove_snapshot(struct snapshot *s, char *why)
+static int exec_rm(void)
{
- int fds[3] = {0, 0, 0};
- assert(!rm_pid);
+ struct snapshot *s = snapshot_currently_being_removed;
char *new_name = being_deleted_name(s);
- int ret = dss_rename(s->name, new_name);
- 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);
+
+ 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 snapshot %s (interval = %i)\n",
- why, s->name, s->interval);
- ret = dss_exec(&rm_pid, argv[0], argv, fds);
+ dss_exec(&remove_pid, argv[0], argv);
+ snapshot_removal_status = HS_RUNNING;
out:
free(new_name);
return ret;
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;
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;
prev = s;
}
assert(victim);
- if (conf.dry_run_given) {
- dss_msg("%s would be removed (interval = %i)\n",
- victim->name, victim->interval);
- continue;
- }
- ret = remove_snapshot(victim, "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 = remove_snapshot(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)
{
- 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;
- return remove_snapshot(s, "oldest");
+ 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)
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;
- if (!low_disk_space && conf.keep_redundant_given)
+ 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;
+ }
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:
+ pre_remove_hook(victim, why);
out:
free_snapshot_list(&sl);
return ret;
}
-static int pre_create_hook(void)
+static void post_create_hook(void)
{
- int ret, fds[3] = {0, 0, 0};
-
- if (!conf.pre_create_hook_given) {
- snapshot_creation_status = SCS_PRE_HOOK_SUCCESS;
- return 0;
- }
- DSS_NOTICE_LOG("executing %s\n", conf.pre_create_hook_arg);
- ret = dss_exec_cmdline_pid(&pre_create_hook_pid,
- conf.pre_create_hook_arg, fds);
- if (ret < 0)
- return ret;
- snapshot_creation_status = SCS_PRE_HOOK_RUNNING;
- return ret;
+ 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);
+ dss_exec_cmdline_pid(&create_pid, cmd);
+ free(cmd);
+ snapshot_creation_status = HS_POST_RUNNING;
}
-static int post_create_hook(void)
+static void post_remove_hook(void)
{
- int ret, fds[3] = {0, 0, 0};
char *cmd;
+ struct snapshot *s = snapshot_currently_being_removed;
- if (!conf.post_create_hook_given) {
- snapshot_creation_status = SCS_READY;
- compute_next_snapshot_time();
- return 0;
- }
- cmd = make_message("%s %s/%s", conf.post_create_hook_arg,
- conf.dest_dir_arg, path_to_last_complete_snapshot);
+ 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(&post_create_hook_pid, cmd, fds);
+ dss_exec_cmdline_pid(&remove_pid, cmd);
free(cmd);
- if (ret < 0)
- return ret;
- snapshot_creation_status = SCS_POST_HOOK_RUNNING;
- return ret;
+ snapshot_removal_status = HS_POST_RUNNING;
}
-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_rsync_process(void)
+static void stop_create_process(void)
{
- if (!rsync_pid || rsync_stopped)
+ if (create_process_stopped)
return;
- kill(SIGSTOP, rsync_pid);
- rsync_stopped = 1;
+ dss_kill(create_pid, SIGSTOP, "suspending create process");
+ create_process_stopped = 1;
}
-static void restart_rsync_process(void)
+static void restart_create_process(void)
{
- if (!rsync_pid || !rsync_stopped)
+ if (!create_process_stopped)
return;
- kill (SIGCONT, rsync_pid);
- rsync_stopped = 0;
+ dss_kill(create_pid, SIGCONT, "resuming create process");
+ create_process_stopped = 0;
}
-
/**
* Print a log message about the exit status of a child.
*/
}
}
/* 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);
return ret;
}
+static void handle_pre_remove_exit(int status)
+{
+ if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
+ snapshot_removal_status = HS_READY;
+ gettimeofday(&next_removal_check, NULL);
+ next_removal_check.tv_sec += 60;
+ return;
+ }
+ snapshot_removal_status = HS_PRE_SUCCESS;
+}
+
static int handle_rm_exit(int status)
{
- rm_pid = 0;
- if (!WIFEXITED(status))
+ if (!WIFEXITED(status)) {
+ snapshot_removal_status = HS_READY;
return -E_INVOLUNTARY_EXIT;
- if (WEXITSTATUS(status))
+ }
+ if (WEXITSTATUS(status)) {
+ snapshot_removal_status = HS_READY;
return -E_BAD_EXIT_CODE;
+ }
+ snapshot_removal_status = HS_SUCCESS;
return 1;
}
-static int wait_for_rm_process(void)
+static void handle_post_remove_exit(void)
+{
+ snapshot_removal_status = HS_READY;
+}
+
+static int handle_remove_exit(int status)
{
- int status, ret = wait_for_process(rm_pid, &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_rm_exit(status);
+ 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)rsync_pid);
+ DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
ret = -E_INVOLUNTARY_EXIT;
- snapshot_creation_status = SCS_READY;
- compute_next_snapshot_time();
+ snapshot_creation_status = HS_READY;
goto out;
}
es = WEXITSTATUS(status);
- if (es == 13) { /* Errors with program diagnostics */
+ /*
+ * 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)rsync_pid, es);
- snapshot_creation_status = SCS_RSYNC_NEEDS_RESTART;
- gettimeofday(&next_snapshot_time, NULL);
- next_snapshot_time.tv_sec += 60;
+ (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)rsync_pid, es);
+ DSS_ERROR_LOG("rsync process %d returned %d\n", (int)create_pid, es);
ret = -E_BAD_EXIT_CODE;
- snapshot_creation_status = SCS_READY;
- compute_next_snapshot_time();
+ snapshot_creation_status = HS_READY;
goto out;
}
ret = rename_incomplete_snapshot(current_snapshot_creation_time);
if (ret < 0)
goto out;
- snapshot_creation_status = SCS_RSYNC_SUCCESS;
+ snapshot_creation_status = HS_SUCCESS;
+ free(name_of_reference_snapshot);
+ name_of_reference_snapshot = NULL;
out:
- rsync_pid = 0;
- rsync_stopped = 0;
+ create_process_stopped = 0;
return ret;
}
static int handle_pre_create_hook_exit(int status)
{
int es, ret;
+ static int warn_count;
if (!WIFEXITED(status)) {
- snapshot_creation_status = SCS_READY;
- compute_next_snapshot_time();
+ snapshot_creation_status = HS_READY;
ret = -E_INVOLUNTARY_EXIT;
goto out;
}
es = WEXITSTATUS(status);
if (es) {
- snapshot_creation_status = SCS_READY;
- compute_next_snapshot_time();
- ret = -E_BAD_EXIT_CODE;
+ 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;
}
- snapshot_creation_status = SCS_PRE_HOOK_SUCCESS;
+ warn_count = 0;
+ snapshot_creation_status = HS_PRE_SUCCESS;
ret = 1;
out:
- pre_create_hook_pid = 0;
return ret;
}
if (ret <= 0)
return ret;
- if (pid == rsync_pid)
- return handle_rsync_exit(status);
- if (pid == rm_pid)
- return handle_rm_exit(status);
- if (pid == pre_create_hook_pid)
- return handle_pre_create_hook_exit(status);
- if (pid == post_create_hook_pid) {
- snapshot_creation_status = SCS_READY;
- compute_next_snapshot_time();
- return 1;
+
+ 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;
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;
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,
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);
}
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)
{
int ret;
- DSS_NOTICE_LOG("SIGHUP\n");
+ 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();
}
switch (sig) {
case SIGINT:
case SIGTERM:
- restart_rsync_process();
- kill_process(rsync_pid);
- kill_process(rm_pid);
+ restart_create_process();
+ dss_kill(create_pid, SIGTERM, NULL);
+ dss_kill(remove_pid, SIGTERM, NULL);
ret = -E_SIGNAL;
break;
case SIGHUP:
return 1;
}
+static int rename_resume_snap(int64_t creation_time)
+{
+ struct snapshot_list sl = {.num_snapshots = 0};
+ 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) /* no snapshots at all */
+ goto out;
+ /* re-use last snapshot if it is incomplete */
+ why = "aborted";
+ if ((s->flags & SS_COMPLETE) == 0)
+ goto out;
+ 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);
+ free_snapshot_list(&sl);
+ return ret;
+}
+
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 *));
(*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);
static int create_snapshot(char **argv)
{
- int ret, fds[3] = {0, 0, 0};
- char *name;
+ int ret;
- name = incomplete_name(current_snapshot_creation_time);
- DSS_NOTICE_LOG("creating new snapshot %s\n", name);
- free(name);
- ret = dss_exec(&rsync_pid, argv[0], argv, fds);
+ ret = rename_resume_snap(current_snapshot_creation_time);
if (ret < 0)
return ret;
- snapshot_creation_status = SCS_RSYNC_RUNNING;
+ dss_exec(&create_pid, argv[0], argv);
+ snapshot_creation_status = HS_RUNNING;
return ret;
}
for (;;) {
fd_set rfds;
- int low_disk_space;
- struct timeval now, *tvp;
+ struct timeval *tvp;
- if (rm_pid)
- tvp = NULL; /* sleep until rm process dies */
+ if (remove_pid)
+ tvp = NULL; /* sleep until rm hook/process dies */
else { /* sleep one minute */
tv.tv_sec = 60;
tv.tv_usec = 0;
}
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;
if (ret < 0)
goto out;
}
- if (rm_pid)
+ if (remove_pid)
continue;
- ret = disk_space_low();
- if (ret < 0)
- goto out;
- low_disk_space = ret;
- ret = try_to_free_disk_space(low_disk_space);
+ if (snapshot_removal_status == HS_PRE_SUCCESS) {
+ ret = exec_rm();
+ if (ret < 0)
+ goto out;
+ continue;
+ }
+ if (snapshot_removal_status == HS_SUCCESS) {
+ post_remove_hook();
+ continue;
+ }
+ ret = try_to_free_disk_space();
if (ret < 0)
goto out;
- if (rm_pid) {
- stop_rsync_process();
+ if (snapshot_removal_status != HS_READY) {
+ stop_create_process();
continue;
}
- restart_rsync_process();
- gettimeofday(&now, NULL);
- if (tv_diff(&next_snapshot_time, &now, NULL) > 0)
- continue;
+ restart_create_process();
switch (snapshot_creation_status) {
- case SCS_READY:
- ret = pre_create_hook();
- if (ret < 0)
- goto out;
+ case HS_READY:
+ if (!next_snapshot_is_due())
+ continue;
+ pre_create_hook();
continue;
- case SCS_PRE_HOOK_RUNNING:
+ case HS_PRE_RUNNING:
+ case HS_RUNNING:
+ case HS_POST_RUNNING:
continue;
- case SCS_RSYNC_NEEDS_RESTART:
+ case HS_PRE_SUCCESS:
+ if (!name_of_reference_snapshot) {
+ free_rsync_argv(rsync_argv);
+ create_rsync_argv(&rsync_argv, ¤t_snapshot_creation_time);
+ }
ret = create_snapshot(rsync_argv);
if (ret < 0)
goto out;
continue;
- case SCS_PRE_HOOK_SUCCESS:
- free_rsync_argv(rsync_argv);
- create_rsync_argv(&rsync_argv, ¤t_snapshot_creation_time);
+ case HS_NEEDS_RESTART:
+ if (!next_snapshot_is_due())
+ continue;
ret = create_snapshot(rsync_argv);
if (ret < 0)
goto out;
continue;
- case SCS_RSYNC_RUNNING:
- continue;
- case SCS_RSYNC_SUCCESS:
- ret = post_create_hook();
- if (ret < 0)
- goto out;
- continue;
- case SCS_POST_HOOK_RUNNING:
+ case HS_SUCCESS:
+ post_create_hook();
continue;
}
}
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)
ret = install_sighandler(SIGHUP);
if (ret < 0)
return ret;
- compute_next_snapshot_time();
ret = select_loop();
if (ret >= 0) /* impossible */
ret = -E_BUG;
{
int ret;
struct snapshot_list sl;
+ struct snapshot *victim;
struct disk_space ds;
+ const char *why;
ret = get_disk_space(".", &ds);
if (ret < 0)
return ret;
log_disk_space(&ds);
- for (;;) {
- dss_get_snapshot_list(&sl);
- ret = remove_outdated_snapshot(&sl);
- free_snapshot_list(&sl);
- if (ret < 0)
- return ret;
- if (!ret)
- break;
- ret = wait_for_rm_process();
- if (ret < 0)
- goto out;
+ dss_get_snapshot_list(&sl);
+ why = "outdated";
+ victim = find_outdated_snapshot(&sl);
+ if (victim)
+ goto rm;
+ why = "redundant";
+ victim = find_redundant_snapshot(&sl);
+ if (victim)
+ goto rm;
+ ret = 0;
+ goto out;
+rm:
+ if (conf.dry_run_given) {
+ dss_msg("%s snapshot %s (interval = %i)\n",
+ why, victim->name, victim->interval);
+ ret = 0;
+ goto out;
}
- for (;;) {
- dss_get_snapshot_list(&sl);
- ret = remove_redundant_snapshot(&sl);
- free_snapshot_list(&sl);
- if (ret < 0)
- return ret;
- if (!ret)
- break;
- ret = wait_for_rm_process();
+ pre_remove_hook(victim, why);
+ if (snapshot_removal_status == HS_PRE_RUNNING) {
+ ret = wait_for_remove_process();
if (ret < 0)
goto out;
+ if (snapshot_removal_status != HS_PRE_SUCCESS)
+ goto out;
}
- return 1;
+ ret = exec_rm();
+ if (ret < 0)
+ goto out;
+ ret = wait_for_remove_process();
+ if (ret < 0)
+ goto out;
+ if (snapshot_removal_status != HS_SUCCESS)
+ goto out;
+ post_remove_hook();
+ 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;
}
free(msg);
return 1;
}
- ret = pre_create_hook();
- if (ret < 0)
- return ret;
- if (pre_create_hook_pid) {
- ret = wait_for_process(pre_create_hook_pid, &status);
+ pre_create_hook();
+ if (create_pid) {
+ ret = wait_for_process(create_pid, &status);
if (ret < 0)
return ret;
ret = handle_pre_create_hook_exit(status);
- if (ret < 0)
+ if (ret <= 0) /* error, or pre-create failed */
return ret;
}
create_rsync_argv(&rsync_argv, ¤t_snapshot_creation_time);
ret = create_snapshot(rsync_argv);
if (ret < 0)
goto out;
- ret = wait_for_process(rsync_pid, &status);
+ ret = wait_for_process(create_pid, &status);
if (ret < 0)
goto out;
ret = handle_rsync_exit(status);
if (ret < 0)
goto out;
post_create_hook();
- if (post_create_hook_pid)
- ret = wait_for_process(post_create_hook_pid, &status);
+ if (create_pid)
+ ret = wait_for_process(create_pid, &status);
out:
free_rsync_argv(rsync_argv);
return ret;
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, ¶ms); /* aborts on errors */
+ }
if (conf.daemon_given)
daemon_init();
ret = change_to_dest_dir();
if (ret < 0)
goto out;
+ dump_dss_config("startup");
ret = setup_signal_handling();
if (ret < 0)
goto out;
projects / dss.git / blobdiff