/*
- * Copyright (C) 2008-2011 Andre Noll <maan@systemlinux.org>
+ * Copyright (C) 2008-2011 Andre Noll <maan@tuebingen.mpg.de>
*
* Licensed under the GPL v2. For licencing details see COPYING.
*/
#include "gcc-compat.h"
#include "cmdline.h"
#include "log.h"
-#include "string.h"
-#include "error.h"
-#include "fd.h"
+#include "str.h"
+#include "err.h"
+#include "file.h"
#include "exec.h"
#include "daemon.h"
-#include "signal.h"
+#include "sig.h"
#include "df.h"
-#include "time.h"
+#include "tv.h"
#include "snap.h"
#include "ipc.h"
static pid_t create_pid;
/** Whether the pre-create-hook/rsync/post-create-hook is currently stopped. */
static int create_process_stopped;
+/** How many times in a row the rsync command failed. */
+static int num_consecutive_rsync_errors;
/** Process id of current pre-remove/rm/post-remove process. */
static pid_t remove_pid;
/** When the next snapshot is due. */
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");
+ 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));
+ DSS_ERROR_LOG(("can not get free disk space: %s\n",
+ dss_strerror(-ret)));
/* we continue on errors from get_disk_space */
"reference_snapshot: %s\n"
"snapshot_creation_status: %s\n"
"snapshot_removal_status: %s\n"
+ "num_consecutive_rsync_errors: %d\n"
,
(int) getpid(),
logfile? conf.logfile_arg : "stderr",
name_of_reference_snapshot?
name_of_reference_snapshot : "(none)",
hook_status_description[snapshot_creation_status],
- hook_status_description[snapshot_removal_status]
+ hook_status_description[snapshot_removal_status],
+ num_consecutive_rsync_errors
);
if (create_pid != 0)
fprintf(log,
static int call_command_handler(void)
{
COMMANDS
- DSS_EMERG_LOG("BUG: did not find command handler\n");
+ DSS_EMERG_LOG(("BUG: did not find command handler\n"));
return -E_BUG;
}
#undef COMMAND
#undef COMMANDS
+static int loglevel = -1;
+static const char *location_file = NULL;
+static int location_line = -1;
+static const char *location_func = NULL;
+
+void dss_log_set_params(int ll, const char *file, int line, const char *func)
+{
+ loglevel = ll;
+ location_file = file;
+ location_line = line;
+ location_func = func;
+}
+
/**
* The log function of dss.
*
*
* All DSS_XXX_LOG() macros use this function.
*/
-__printf_2_3 void dss_log(int ll, const char* fmt,...)
+__printf_1_2 void dss_log(const char* fmt,...)
{
va_list argp;
FILE *outfd;
time_t t1;
char str[255] = "";
- if (ll < conf.loglevel_arg)
+ if (loglevel < conf.loglevel_arg)
return;
outfd = logfile? logfile : stderr;
time(&t1);
strftime(str, sizeof(str), "%b %d %H:%M:%S", tm);
fprintf(outfd, "%s ", str);
if (conf.loglevel_arg <= INFO)
- fprintf(outfd, "%i: ", ll);
+ fprintf(outfd, "%i: ", loglevel);
+#ifdef DSS_NO_FUNC_NAMES
+ fprintf(outfd, "%s:%d: ", location_file, location_line);
+#else
+ fprintf(outfd, "%s: ", location_func);
+#endif
va_start(argp, fmt);
vfprintf(outfd, fmt, argp);
va_end(argp);
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");
+ DSS_DEBUG_LOG(("next snapshot: now\n"));
return 1;
}
- DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n",
- next_snapshot_time - now);
+ DSS_DEBUG_LOG(("next snapshot due in %" PRId64 " seconds\n",
+ next_snapshot_time - now));
return 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);
+ 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;
}
if (!s)
return;
- DSS_DEBUG_LOG("%s snapshot %s\n", why, s->name);
+ 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);
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_DEBUG_LOG(("executing %s\n", cmd));
dss_exec_cmdline_pid(&remove_pid, cmd);
free(cmd);
snapshot_removal_status = HS_PRE_RUNNING;
assert(snapshot_removal_status == HS_PRE_SUCCESS);
assert(remove_pid == 0);
- DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
+ DSS_NOTICE_LOG(("removing %s (interval = %i)\n", s->name, s->interval));
ret = dss_rename(s->name, new_name);
if (ret < 0)
goto out;
struct snapshot *s;
int i;
- DSS_DEBUG_LOG("looking for orphaned snapshots\n");
+ DSS_DEBUG_LOG(("looking for orphaned snapshots\n"));
FOR_EACH_SNAPSHOT(s, i, sl) {
if (snapshot_is_being_created(s))
continue;
struct snapshot *s;
unsigned missing = 0;
- DSS_DEBUG_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 = desired_number_of_snapshots(interval, conf.num_intervals_arg);
unsigned num = sl->interval_count[interval];
int i;
struct snapshot *s;
- DSS_DEBUG_LOG("looking for snapshots belonging to intervals >= %d\n",
- conf.num_intervals_arg);
+ 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;
return NULL;
}
-struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
+static struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
{
- int i;
- struct snapshot *s;
+ int i, num_complete;
+ struct snapshot *s, *ref = NULL;
+
+ num_complete = num_complete_snapshots(sl);
+ if (num_complete <= conf.min_complete_arg)
+ return NULL;
FOR_EACH_SNAPSHOT(s, i, sl) {
if (snapshot_is_being_created(s))
continue;
- if (is_reference_snapshot(s))
+ if (is_reference_snapshot(s)) { /* avoid this one */
+ ref = s;
continue;
- DSS_INFO_LOG("oldest removable snapshot: %s\n", s->name);
+ }
+ DSS_INFO_LOG(("oldest removable snapshot: %s\n", s->name));
return s;
}
- return NULL;
+ assert(ref);
+ DSS_WARNING_LOG(("removing reference snapshot %s\n", ref->name));
+ return ref;
}
static int rename_incomplete_snapshot(int64_t start)
{
char *old_name;
int ret;
+ int64_t now;
+ /*
+ * We don't want the dss_rename() below to fail with EEXIST because the
+ * last complete snapshot was created (and completed) in the same
+ * second as this one.
+ */
+ while ((now = get_current_time()) == start)
+ sleep(1);
free(path_to_last_complete_snapshot);
- ret = complete_name(start, get_current_time(),
- &path_to_last_complete_snapshot);
+ ret = complete_name(start, now, &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);
+ DSS_NOTICE_LOG(("%s -> %s\n", old_name,
+ path_to_last_complete_snapshot));
free(old_name);
return ret;
}
if (next_snapshot_is_due())
return 0;
}
+ /*
+ * Idle and --keep_redundant not given, or low disk space. Look at
+ * existing snapshots.
+ */
dss_get_snapshot_list(&sl);
ret = 0;
- if (!low_disk_space && sl.num_snapshots <= 1)
+ /*
+ * 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 << conf.num_intervals_arg)
goto out;
why = "outdated";
victim = find_outdated_snapshot(&sl);
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");
+ /* try harder only if disk space is low */
+ if (!low_disk_space)
+ goto out;
+ 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");
+ DSS_CRIT_LOG(("uhuhu: disk space low and nothing to remove\n"));
ret = -ERRNO_TO_DSS_ERROR(ENOSPC);
goto out;
remove:
{
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_NOTICE_LOG(("executing %s\n", cmd));
dss_exec_cmdline_pid(&create_pid, cmd);
free(cmd);
snapshot_creation_status = HS_POST_RUNNING;
cmd = make_message("%s %s/%s", conf.post_remove_hook_arg,
conf.dest_dir_arg, s->name);
- DSS_NOTICE_LOG("executing %s\n", cmd);
+ DSS_NOTICE_LOG(("executing %s\n", cmd));
dss_exec_cmdline_pid(&remove_pid, cmd);
free(cmd);
snapshot_removal_status = HS_POST_RUNNING;
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);
+ 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);
+ 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)
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));
+ 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));
+ 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);
+ DSS_WARNING_LOG(("child %i terminated abormally\n", (int)pid));
}
static int wait_for_process(pid_t pid, int *status)
{
int ret;
- DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
+ DSS_DEBUG_LOG(("Waiting for process %d to terminate\n", (int)pid));
for (;;) {
fd_set rfds;
dss_kill(pid, SIGTERM, "killing child process");
}
if (ret < 0)
- DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
+ DSS_ERROR_LOG(("failed to wait for process %d\n", (int)pid));
else
log_termination_msg(pid, *status);
return ret;
int es, ret;
if (!WIFEXITED(status)) {
- DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)create_pid);
+ DSS_ERROR_LOG(("rsync process %d died involuntary\n", (int)create_pid));
ret = -E_INVOLUNTARY_EXIT;
snapshot_creation_status = HS_READY;
goto out;
es = WEXITSTATUS(status);
/*
* Restart rsync on non-fatal errors:
- * 12: Error in rsync protocol data stream
- * 13: Errors with program diagnostics
+ * 24: Partial transfer due to vanished source files
*/
- if (es == 12 || es == 13) {
- DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
- (int)create_pid, es);
+ if (es != 0 && es != 24) {
+ DSS_WARNING_LOG(("rsync exit code %d, error count %d\n",
+ es, ++num_consecutive_rsync_errors));
+ if (conf.create_given) {
+ ret = -E_BAD_EXIT_CODE;
+ goto out;
+ }
+ if (num_consecutive_rsync_errors > conf.max_rsync_errors_arg) {
+ ret = -E_TOO_MANY_RSYNC_ERRORS;
+ snapshot_creation_status = HS_READY;
+ goto out;
+ }
+ DSS_WARNING_LOG(("restarting rsync process\n"));
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;
- }
+ num_consecutive_rsync_errors = 0;
ret = rename_incomplete_snapshot(current_snapshot_creation_time);
if (ret < 0)
goto out;
es = WEXITSTATUS(status);
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");
+ 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;
ret = 1;
break;
default:
- DSS_EMERG_LOG("BUG: create can't die in status %d\n",
- snapshot_creation_status);
+ DSS_EMERG_LOG(("BUG: create can't die in status %d\n",
+ snapshot_creation_status));
return -E_BUG;
}
create_pid = 0;
return ret;
return ret;
}
- DSS_EMERG_LOG("BUG: unknown process %d died\n", (int)pid);
+ DSS_EMERG_LOG(("BUG: unknown process %d died\n", (int)pid));
return -E_BUG;
}
static int check_config(void)
{
if (conf.unit_interval_arg <= 0) {
- DSS_ERROR_LOG("bad unit interval: %i\n", conf.unit_interval_arg);
+ 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);
+ DSS_DEBUG_LOG(("unit interval: %i day(s)\n", conf.unit_interval_arg));
+ if (conf.num_intervals_arg <= 0 || conf.num_intervals_arg > 30) {
+ 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);
+ DSS_DEBUG_LOG(("number of intervals: %i\n", conf.num_intervals_arg));
return 1;
}
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);
+ DSS_ERROR_LOG(("failed to stat config file %s\n", config_file));
goto out;
}
if (config_file_exists) {
logfile = open_log(conf.logfile_arg);
log_welcome(conf.loglevel_arg);
}
- DSS_DEBUG_LOG("loglevel: %d\n", 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));
+ DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
return ret;
}
static int change_to_dest_dir(void)
{
- DSS_INFO_LOG("changing cwd to %s\n", conf.dest_dir_arg);
+ DSS_INFO_LOG(("changing cwd to %s\n", conf.dest_dir_arg));
return dss_chdir(conf.dest_dir_arg);
}
{
int ret;
- DSS_NOTICE_LOG("SIGHUP, re-reading config\n");
+ DSS_NOTICE_LOG(("SIGHUP, re-reading config\n"));
dump_dss_config("old");
ret = parse_config_file(1);
if (ret < 0)
return change_to_dest_dir();
}
+static void kill_children(void)
+{
+ restart_create_process();
+ dss_kill(create_pid, SIGTERM, NULL);
+ dss_kill(remove_pid, SIGTERM, NULL);
+}
+
static int handle_signal(void)
{
int sig, ret = next_signal();
switch (sig) {
case SIGINT:
case SIGTERM:
- restart_create_process();
- dss_kill(create_pid, SIGTERM, NULL);
- dss_kill(remove_pid, SIGTERM, NULL);
+ kill_children();
ret = -E_SIGNAL;
break;
case SIGHUP:
}
out:
if (ret < 0)
- DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
+ DSS_ERROR_LOG(("%s\n", dss_strerror(-ret)));
return ret;
}
s = find_orphaned_snapshot(&sl);
out:
if (s) {
- DSS_INFO_LOG("reusing %s snapshot %s\n", why, s->name);
+ 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);
+ DSS_NOTICE_LOG(("creating new snapshot %s\n", new_name));
free(new_name);
free_snapshot_list(&sl);
return ret;
*argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
(*argv)[i++] = dss_strdup("rsync");
- (*argv)[i++] = dss_strdup("-aq");
+ (*argv)[i++] = dss_strdup("-a");
(*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);
+ 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");
+ 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);
(*argv)[i++] = incomplete_name(*num);
(*argv)[i++] = NULL;
for (j = 0; j < i; j++)
- DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
+ DSS_DEBUG_LOG(("argv[%d] = %s\n", j, (*argv)[j]));
}
static void free_rsync_argv(char **argv)
argv[1] = dss_strerror(-exit_code);
argv[2] = NULL;
- DSS_NOTICE_LOG("executing %s %s\n", argv[0], argv[1]);
+ DSS_NOTICE_LOG(("executing %s %s\n", argv[0], argv[1]));
dss_exec(&pid, conf.exit_hook_arg, argv);
}
free(config_file);
if (ret < 0) {
- DSS_EMERG_LOG("failed to lock: %s\n", dss_strerror(-ret));
+ DSS_EMERG_LOG(("failed to lock: %s\n", dss_strerror(-ret)));
exit(EXIT_FAILURE);
}
}
lock_dss_or_die();
if (conf.dry_run_given) {
- DSS_ERROR_LOG("dry_run not supported by this command\n");
+ DSS_ERROR_LOG(("dry_run not supported by this command\n"));
return -E_SYNTAX;
}
ret = install_sighandler(SIGHUP);
ret = select_loop();
if (ret >= 0) /* impossible */
ret = -E_BUG;
+ kill_children();
exit_hook(ret);
return ret;
}
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;
}
{
int ret;
- DSS_INFO_LOG("setting up signal handlers\n");
+ DSS_INFO_LOG(("setting up signal handlers\n"));
signal_pipe = signal_init(); /* always successful */
ret = install_sighandler(SIGINT);
if (ret < 0)
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;
if (ret < 0)
goto out;
ret = call_command_handler();
+ signal_shutdown();
out:
if (ret < 0)
- DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
+ DSS_EMERG_LOG(("%s\n", dss_strerror(-ret)));
exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
}
projects / dss.git / blobdiff