+/*
+ * Copyright (C) 2008-2011 Andre Noll <maan@systemlinux.org>
+ *
+ * Licensed under the GPL v2. For licencing details see COPYING.
+ */
#include <string.h>
#include <stdlib.h>
#include <stdarg.h>
#include "signal.h"
#include "df.h"
#include "time.h"
+#include "snap.h"
-
-struct gengetopt_args_info conf;
-char *dss_error_txt = NULL;
+/** Command line and config file options. */
+static struct gengetopt_args_info conf;
+/** Non-NULL if we log to a file. */
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. */
-struct timeval next_snapshot_time;
-/** The pid of the pre-create hook. */
-pid_t pre_create_hook_pid;
-/** The pid of the post-create hook. */
-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. */
+enum hook_status snapshot_creation_status;
+/** \sa \ref snap.h for details. */
+enum hook_status snapshot_removal_status;
-/* Creation time of the snapshot currently being created. */
-int64_t current_snapshot_creation_time;
-static char *path_to_last_complete_snapshot;
+DEFINE_DSS_ERRLIST;
+static const char *hook_status_description[] = {HOOK_STATUS_ARRAY};
-enum {
- /** We are ready to take the next snapshot. */
- SCS_READY,
- /** The pre-creation hook has been started. */
- SCS_PRE_HOOK_RUNNING,
- /** The pre-creation hook exited successfully. */
- SCS_PRE_HOOK_SUCCESS,
- /** The rsync process is running. */
- SCS_RSYNC_RUNNING,
- /** The rsync process exited successfully. */
- SCS_RSYNC_SUCCESS,
- /** The post-create hook has been started- */
- SCS_POST_HOOK_RUNNING,
-};
+/* 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 unsigned snapshot_creation_status;
+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;
-DEFINE_DSS_ERRLIST;
+ 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 \
COMMAND(create) \
COMMAND(prune) \
COMMAND(run)
-#define COMMAND(x) int com_ ##x(void);
+#define COMMAND(x) static int com_ ##x(void);
COMMANDS
#undef COMMAND
#define COMMAND(x) if (conf.x ##_given) return com_ ##x();
-int call_command_handler(void)
+static int call_command_handler(void)
{
COMMANDS
DSS_EMERG_LOG("BUG: did not find command handler\n");
- exit(EXIT_FAILURE);
+ return -E_BUG;
}
#undef COMMAND
#undef COMMANDS
-/*
- * complete, not being deleted: 1204565370-1204565371.Sun_Mar_02_2008_14_33-Sun_Mar_02_2008_14_43
- * complete, being deleted: 1204565370-1204565371.being_deleted
- * incomplete, not being deleted: 1204565370-incomplete
- * incomplete, being deleted: 1204565370-incomplete.being_deleted
- */
-enum snapshot_status_flags {
- /** The rsync process terminated successfully. */
- SS_COMPLETE = 1,
- /** The rm process is running to remove this snapshot. */
- SS_BEING_DELETED = 2,
-};
-
-struct snapshot {
- char *name;
- int64_t creation_time;
- int64_t completion_time;
- enum snapshot_status_flags flags;
- unsigned interval;
-};
-
-/*
- * An edge snapshot is either the oldest one or the newest one.
+/**
+ * The log function of dss.
*
- * We need to find either of them occasionally: The create code
- * needs to know the newest snapshot because that is the one
- * used as the link destination dir. The pruning code needs to
- * find the oldest one in case disk space becomes low.
+ * \param ll Loglevel.
+ * \param fml Usual format string.
+ *
+ * All DSS_XXX_LOG() macros use this function.
*/
-struct edge_snapshot_data {
- int64_t now;
- struct snapshot snap;
-};
-
__printf_2_3 void dss_log(int ll, const char* fmt,...)
{
va_list argp;
/**
* Print a message either to stdout or to the log file.
*/
-__printf_1_2 void dss_msg(const char* fmt,...)
+static __printf_1_2 void dss_msg(const char* fmt,...)
{
FILE *outfd = conf.daemon_given? logfile : stdout;
va_list argp;
va_end(argp);
}
-/**
- * Return the desired number of snapshots of an interval.
- */
-unsigned num_snapshots(int interval)
+static void dss_get_snapshot_list(struct snapshot_list *sl)
{
- unsigned n;
-
- assert(interval >= 0);
-
- if (interval >= conf.num_intervals_arg)
- return 0;
- n = conf.num_intervals_arg - interval - 1;
- return 1 << n;
+ get_snapshot_list(sl, conf.unit_interval_arg, conf.num_intervals_arg);
}
-/* return: Whether dirname is a snapshot directory (0: no, 1: yes) */
-int is_snapshot(const char *dirname, int64_t now, struct snapshot *s)
+static int64_t compute_next_snapshot_time(void)
{
- int i, ret;
- char *dash, *dot, *tmp;
- int64_t num;
+ 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;
+ struct snapshot *s = NULL;
+ struct snapshot_list sl;
- assert(dirname);
- dash = strchr(dirname, '-');
- if (!dash || !dash[1] || dash == dirname)
- return 0;
- for (i = 0; dirname[i] != '-'; i++)
- if (!isdigit(dirname[i]))
- return 0;
- tmp = dss_strdup(dirname);
- tmp[i] = '\0';
- ret = dss_atoi64(tmp, &num);
- free(tmp);
- if (ret < 0) {
- free(dss_error_txt);
- return 0;
- }
- assert(num >= 0);
- if (num > now)
- return 0;
- s->creation_time = num;
- //DSS_DEBUG_LOG("%s start time: %lli\n", dirname, (long long)s->creation_time);
- s->interval = (long long) ((now - s->creation_time)
- / conf.unit_interval_arg / 24 / 3600);
- if (!strcmp(dash + 1, "incomplete")) {
- s->completion_time = -1;
- s->flags = 0; /* neither complete, nor being deleted */
- goto success;
- }
- if (!strcmp(dash + 1, "incomplete.being_deleted")) {
- s->completion_time = -1;
- s->flags = SS_BEING_DELETED; /* mot cpmplete, being deleted */
- goto success;
- }
- tmp = dash + 1;
- dot = strchr(tmp, '.');
- if (!dot || !dot[1] || dot == tmp)
- return 0;
- for (i = 0; tmp[i] != '.'; i++)
- if (!isdigit(tmp[i]))
- return 0;
- tmp = dss_strdup(dash + 1);
- tmp[i] = '\0';
- ret = dss_atoi64(tmp, &num);
- free(tmp);
- if (ret < 0) {
- free(dss_error_txt);
- return 0;
+ dss_get_snapshot_list(&sl);
+ FOR_EACH_SNAPSHOT(s, i, &sl) {
+ if (!(s->flags & SS_COMPLETE))
+ continue;
+ num_complete_snapshots++;
+ x += s->completion_time - s->creation_time;
}
- if (num > now)
- return 0;
- s->completion_time = num;
- s->flags = SS_COMPLETE;
- if (!strcmp(dot + 1, "being_deleted"))
- s->flags |= SS_BEING_DELETED;
-success:
- s->name = dss_strdup(dirname);
- return 1;
-}
-
-int64_t get_current_time(void)
-{
- time_t now;
- time(&now);
- DSS_DEBUG_LOG("now: %lli\n", (long long) now);
- return (int64_t)now;
-}
-
-char *incomplete_name(int64_t start)
-{
- return make_message("%lli-incomplete", (long long)start);
-}
+ assert(x >= 0);
-char *being_deleted_name(struct snapshot *s)
-{
- if (s->flags & SS_COMPLETE)
- return make_message("%lli-%lli.being_deleted",
- (long long)s->creation_time,
- (long long)s->completion_time);
- return make_message("%lli-incomplete.being_deleted",
- (long long)s->creation_time);
+ 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);
+ return ret;
}
-int complete_name(int64_t start, int64_t end, char **result)
+static inline void invalidate_next_snapshot_time(void)
{
- struct tm start_tm, end_tm;
- time_t *start_seconds = (time_t *) (uint64_t *)&start; /* STFU, gcc */
- time_t *end_seconds = (time_t *) (uint64_t *)&end; /* STFU, gcc */
- char start_str[200], end_str[200];
-
- if (!localtime_r(start_seconds, &start_tm))
- return -E_LOCALTIME;
- if (!localtime_r(end_seconds, &end_tm))
- return -E_LOCALTIME;
- if (!strftime(start_str, sizeof(start_str), "%a_%b_%d_%Y_%H_%M_%S", &start_tm))
- return -E_STRFTIME;
- if (!strftime(end_str, sizeof(end_str), "%a_%b_%d_%Y_%H_%M_%S", &end_tm))
- return -E_STRFTIME;
- *result = make_message("%lli-%lli.%s-%s", (long long) start, (long long) end,
- start_str, end_str);
- return 1;
+ next_snapshot_time = 0;
}
-struct snapshot_list {
- int64_t now;
- unsigned num_snapshots;
- unsigned array_size;
- struct snapshot **snapshots;
- /**
- * Array of size num_intervals + 1
- *
- * It contains the number of snapshots in each interval. interval_count[num_intervals]
- * is the number of snapshots which belong to any interval greater than num_intervals.
- */
- unsigned *interval_count;
-};
-
-#define FOR_EACH_SNAPSHOT(s, i, sl) \
- for ((i) = 0; (i) < (sl)->num_snapshots && ((s) = (sl)->snapshots[(i)]); (i)++)
-
-#define NUM_COMPARE(x, y) ((int)((x) < (y)) - (int)((x) > (y)))
-
-static int compare_snapshots(const void *a, const void *b)
+static inline int next_snapshot_time_is_valid(void)
{
- struct snapshot *s1 = *(struct snapshot **)a;
- struct snapshot *s2 = *(struct snapshot **)b;
- return NUM_COMPARE(s2->creation_time, s1->creation_time);
+ return next_snapshot_time != 0;
}
-/** Compute the minimum of \a a and \a b. */
-#define DSS_MIN(a,b) ((a) < (b) ? (a) : (b))
-
-int add_snapshot(const char *dirname, void *private)
+static int next_snapshot_is_due(void)
{
- struct snapshot_list *sl = private;
- struct snapshot s;
- int ret = is_snapshot(dirname, sl->now, &s);
+ int64_t now = get_current_time();
- if (!ret)
+ 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;
- if (sl->num_snapshots >= sl->array_size) {
- sl->array_size = 2 * sl->array_size + 1;
- sl->snapshots = dss_realloc(sl->snapshots,
- sl->array_size * sizeof(struct snapshot *));
}
- sl->snapshots[sl->num_snapshots] = dss_malloc(sizeof(struct snapshot));
- *(sl->snapshots[sl->num_snapshots]) = s;
- sl->interval_count[DSS_MIN(s.interval, conf.num_intervals_arg)]++;
- sl->num_snapshots++;
- return 1;
+ DSS_DEBUG_LOG("next snapshot due in %" PRId64 " seconds\n",
+ next_snapshot_time - now);
+ return 0;
}
-void get_snapshot_list(struct snapshot_list *sl)
+static void pre_create_hook(void)
{
- sl->now = get_current_time();
- sl->num_snapshots = 0;
- sl->array_size = 0;
- sl->snapshots = NULL;
- sl->interval_count = dss_calloc((conf.num_intervals_arg + 1) * sizeof(unsigned));
- for_each_subdir(add_snapshot, sl);
- qsort(sl->snapshots, sl->num_snapshots, sizeof(struct snapshot *),
- compare_snapshots);
+ 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;
}
-void free_snapshot_list(struct snapshot_list *sl)
+static void pre_remove_hook(struct snapshot *s, const char *why)
{
- int i;
- struct snapshot *s;
-
- FOR_EACH_SNAPSHOT(s, i, sl) {
- free(s->name);
- free(s);
- }
- free(sl->interval_count);
- sl->interval_count = NULL;
- free(sl->snapshots);
- sl->snapshots = NULL;
- sl->num_snapshots = 0;
-}
+ char *cmd;
-void stop_rsync_process(void)
-{
- if (!rsync_pid || rsync_stopped)
+ if (!s)
return;
- kill(SIGSTOP, rsync_pid);
- rsync_stopped = 1;
+ 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;
}
-void restart_rsync_process(void)
+static int exec_rm(void)
{
- if (!rsync_pid || !rsync_stopped)
- return;
- kill (SIGCONT, rsync_pid);
- rsync_stopped = 0;
+ struct snapshot *s = snapshot_currently_being_removed;
+ char *new_name = being_deleted_name(s);
+ 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_exec(&remove_pid, argv[0], argv);
+ snapshot_removal_status = HS_RUNNING;
+out:
+ free(new_name);
+ return ret;
}
-/**
- * Print a log message about the exit status of a child.
- */
-void log_termination_msg(pid_t pid, int status)
+static int snapshot_is_being_created(struct snapshot *s)
{
- if (WIFEXITED(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));
- else
- DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
+ return s->creation_time == current_snapshot_creation_time;
}
-int wait_for_process(pid_t pid, int *status)
+static struct snapshot *find_orphaned_snapshot(struct snapshot_list *sl)
{
- int ret;
-
- DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
- for (;;) {
- fd_set rfds;
+ struct snapshot *s;
+ int i;
- FD_ZERO(&rfds);
- FD_SET(signal_pipe, &rfds);
- ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
- if (ret < 0)
- break;
- ret = next_signal();
- if (!ret)
+ DSS_DEBUG_LOG("looking for orphaned snapshots\n");
+ FOR_EACH_SNAPSHOT(s, i, sl) {
+ if (snapshot_is_being_created(s))
continue;
- if (ret == SIGCHLD) {
- ret = waitpid(pid, status, 0);
- if (ret >= 0)
- break;
- if (errno != EINTR) { /* error */
- ret = -ERRNO_TO_DSS_ERROR(errno);
- break;
- }
- }
- /* SIGINT or SIGTERM */
- DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
- kill(pid, SIGTERM);
+ /*
+ * 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;
}
- if (ret < 0)
- DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
- else
- log_termination_msg(pid, *status);
- return ret;
+ /* no orphaned snapshots */
+ return NULL;
}
-int remove_snapshot(struct snapshot *s)
+static int is_reference_snapshot(struct snapshot *s)
{
- int fds[3] = {0, 0, 0};
- assert(!rm_pid);
- char *new_name = being_deleted_name(s);
- int ret = dss_rename(s->name, new_name);
- char *argv[] = {"rm", "-rf", new_name, NULL};
-
- if (ret < 0)
- goto out;
- DSS_NOTICE_LOG("removing %s (interval = %i)\n", s->name, s->interval);
- stop_rsync_process();
- ret = dss_exec(&rm_pid, argv[0], argv, fds);
-out:
- free(new_name);
- return ret;
+ 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
*/
-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;
- DSS_INFO_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 = num_snapshots(interval);
+ unsigned keep = desired_number_of_snapshots(interval, conf.num_intervals_arg);
unsigned num = sl->interval_count[interval];
struct snapshot *victim = NULL, *prev = NULL;
int64_t score = LONG_MAX;
FOR_EACH_SNAPSHOT(s, i, sl) {
int64_t this_score;
+ 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);
- return ret < 0? ret : 1;
+ return victim;
}
- return 0;
+ return NULL;
}
-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_INFO_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 (s->interval <= conf.num_intervals_arg)
+ if (snapshot_is_being_created(s))
continue;
- if (conf.dry_run_given) {
- dss_msg("%s would be removed (interval = %i)\n",
- s->name, s->interval);
+ if (is_reference_snapshot(s))
continue;
- }
- ret = remove_snapshot(s);
- if (ret < 0)
- return ret;
- return 1;
+ if (s->interval < conf.num_intervals_arg)
+ continue;
+ return s;
}
- return 0;
+ return NULL;
}
-int handle_rm_exit(int status)
+struct snapshot *find_oldest_removable_snapshot(struct snapshot_list *sl)
{
- rm_pid = 0;
- if (!WIFEXITED(status))
- return -E_INVOLUNTARY_EXIT;
- if (WEXITSTATUS(status))
- return -E_BAD_EXIT_CODE;
- 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;
}
-int wait_for_rm_process(void)
+static int rename_incomplete_snapshot(int64_t start)
{
- int status, ret = wait_for_process(rm_pid, &status);
+ char *old_name;
+ int ret;
+ free(path_to_last_complete_snapshot);
+ ret = complete_name(start, get_current_time(),
+ &path_to_last_complete_snapshot);
if (ret < 0)
return ret;
- return handle_rm_exit(status);
+ 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);
+ free(old_name);
+ return ret;
}
-void kill_process(pid_t pid)
+static int try_to_free_disk_space(void)
{
- if (!pid)
- return;
- DSS_WARNING_LOG("sending SIGTERM to pid %d\n", (int)pid);
- kill(pid, SIGTERM);
+ int ret;
+ struct snapshot_list sl;
+ struct snapshot *victim;
+ struct timeval now;
+ const char *why;
+ int low_disk_space;
+
+ 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 = 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");
+ 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;
}
-int check_config(void)
+static void post_create_hook(void)
{
- if (conf.unit_interval_arg <= 0) {
- 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);
- return -E_INVALID_NUMBER;
- }
- DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
- return 1;
+ 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;
}
-/* exits on errors */
-void parse_config_file(int override)
+static void post_remove_hook(void)
{
- int ret;
- char *config_file;
- struct stat statbuf;
- char *old_logfile_arg = NULL;
- int old_daemon_given = 0;
+ char *cmd;
+ struct snapshot *s = snapshot_currently_being_removed;
- if (conf.config_file_given)
- config_file = dss_strdup(conf.config_file_arg);
- else {
- char *home = get_homedir();
- config_file = make_message("%s/.dssrc", home);
- free(home);
- }
- if (override) { /* SIGHUP */
- if (conf.logfile_given)
- old_logfile_arg = dss_strdup(conf.logfile_arg);
- old_daemon_given = conf.daemon_given;
- }
+ assert(s);
- ret = stat(config_file, &statbuf);
- if (ret && conf.config_file_given) {
- ret = -ERRNO_TO_DSS_ERROR(errno);
- DSS_ERROR_LOG("failed to stat config file %s\n", config_file);
- goto out;
+ 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_exec_cmdline_pid(&remove_pid, cmd);
+ free(cmd);
+ snapshot_removal_status = HS_POST_RUNNING;
+}
+
+static void dss_kill(pid_t pid, int sig, const char *msg)
+{
+ 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 (!ret) {
- struct cmdline_parser_params params = {
- .override = override,
- .initialize = 0,
- .check_required = 1,
- .check_ambiguity = 0
- };
- cmdline_parser_config_file(config_file, &conf, ¶ms);
- }
- ret = check_config();
- if (ret < 0)
- goto out;
- if (override) {
- /* don't change daemon mode on SIGHUP */
- conf.daemon_given = old_daemon_given;
- close_log(logfile);
- logfile = NULL;
- if (conf.logfile_given)
- free(old_logfile_arg);
- else if (conf.daemon_given) { /* re-use old logfile */
- conf.logfile_arg = old_logfile_arg;
- conf.logfile_given = 1;
- }
- }
- if (conf.logfile_given) {
- logfile = open_log(conf.logfile_arg);
- log_welcome(conf.loglevel_arg);
- }
- DSS_EMERG_LOG("loglevel: %d\n", conf.loglevel_arg);
-// cmdline_parser_dump(logfile? logfile : stdout, &conf);
- ret = dss_chdir(conf.dest_dir_arg);
-out:
- free(config_file);
- if (ret >= 0)
+
+ 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_EMERG_LOG("%s\n", dss_strerror(-ret));
- exit(EXIT_FAILURE);
+ DSS_INFO_LOG("failed to send signal %d (%s) to pid %d (%s process)\n",
+ sig, signame, (int)pid, process_name);
}
-void handle_sighup(void)
+static void stop_create_process(void)
{
- DSS_NOTICE_LOG("SIGHUP\n");
- parse_config_file(1);
+ if (create_process_stopped)
+ return;
+ dss_kill(create_pid, SIGSTOP, "suspending create process");
+ create_process_stopped = 1;
}
-int rename_incomplete_snapshot(int64_t start)
+static void restart_create_process(void)
{
- char *old_name;
- int ret;
-
- free(path_to_last_complete_snapshot);
- ret = complete_name(start, get_current_time(),
- &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);
- free(old_name);
- return ret;
+ if (!create_process_stopped)
+ return;
+ dss_kill(create_pid, SIGCONT, "resuming create process");
+ create_process_stopped = 0;
}
-void compute_next_snapshot_time(void)
+/**
+ * Print a log message about the exit status of a child.
+ */
+static void log_termination_msg(pid_t pid, int status)
{
- struct timeval now, unit_interval = {.tv_sec = 24 * 3600 * conf.unit_interval_arg},
- tmp, diff;
- int64_t x = 0;
- unsigned wanted = num_snapshots(0), num_complete_snapshots = 0;
- int i, ret;
- struct snapshot *s = NULL;
- struct snapshot_list sl;
-
- assert(snapshot_creation_status == SCS_READY);
- current_snapshot_creation_time = 0;
- get_snapshot_list(&sl);
- FOR_EACH_SNAPSHOT(s, i, &sl) {
- if (!(s->flags & SS_COMPLETE))
- continue;
- num_complete_snapshots++;
- 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 */
- gettimeofday(&now, NULL);
- if (ret < 0 || !s)
- goto min_sleep;
- tv_divide(wanted, &diff, &tmp); /* sleep time betweeen two snapshots */
- diff.tv_sec = s->completion_time;
- diff.tv_usec = 0;
- tv_add(&diff, &tmp, &next_snapshot_time);
- if (tv_diff(&now, &next_snapshot_time, NULL) < 0)
- goto out;
-min_sleep:
- next_snapshot_time = now;
- next_snapshot_time.tv_sec += 60;
-out:
- free_snapshot_list(&sl);
+ if (WIFEXITED(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));
+ else
+ DSS_WARNING_LOG("child %i terminated abormally\n", (int)pid);
}
-int handle_rsync_exit(int status)
+static int wait_for_process(pid_t pid, int *status)
{
- int es, ret;
+ int ret;
- if (!WIFEXITED(status)) {
- DSS_ERROR_LOG("rsync process %d died involuntary\n", (int)rsync_pid);
- ret = -E_INVOLUNTARY_EXIT;
- snapshot_creation_status = SCS_READY;
- compute_next_snapshot_time();
- goto out;
- }
- es = WEXITSTATUS(status);
- if (es != 0 && es != 23 && es != 24) {
- DSS_ERROR_LOG("rsync process %d returned %d\n", (int)rsync_pid, es);
- ret = -E_BAD_EXIT_CODE;
- snapshot_creation_status = SCS_READY;
- compute_next_snapshot_time();
- goto out;
+ DSS_DEBUG_LOG("Waiting for process %d to terminate\n", (int)pid);
+ for (;;) {
+ fd_set rfds;
+
+ FD_ZERO(&rfds);
+ FD_SET(signal_pipe, &rfds);
+ ret = dss_select(signal_pipe + 1, &rfds, NULL, NULL);
+ if (ret < 0)
+ break;
+ ret = next_signal();
+ if (!ret)
+ continue;
+ if (ret == SIGCHLD) {
+ ret = waitpid(pid, status, 0);
+ if (ret >= 0)
+ break;
+ if (errno != EINTR) { /* error */
+ ret = -ERRNO_TO_DSS_ERROR(errno);
+ break;
+ }
+ }
+ /* SIGINT or SIGTERM */
+ dss_kill(pid, SIGTERM, "killing child process");
}
- ret = rename_incomplete_snapshot(current_snapshot_creation_time);
if (ret < 0)
- goto out;
- snapshot_creation_status = SCS_RSYNC_SUCCESS;
-out:
- rsync_pid = 0;
- rsync_stopped = 0;
+ DSS_ERROR_LOG("failed to wait for process %d\n", (int)pid);
+ else
+ log_termination_msg(pid, *status);
return ret;
}
-int get_newest_complete(const char *dirname, void *private)
+static void handle_pre_remove_exit(int status)
{
- struct edge_snapshot_data *esd = private;
- struct snapshot s;
- int ret = is_snapshot(dirname, esd->now, &s);
-
- if (ret <= 0)
- return 1;
- if (s.flags != SS_COMPLETE) /* incomplete or being deleted */
- return 1;
- if (s.creation_time < esd->snap.creation_time)
- return 1;
- free(esd->snap.name);
- esd->snap = s;
- return 1;
-}
-
-__malloc char *name_of_newest_complete_snapshot(void)
-{
- struct edge_snapshot_data esd = {
- .now = get_current_time(),
- .snap = {.creation_time = -1}
- };
- for_each_subdir(get_newest_complete, &esd);
- return esd.snap.name;
+ 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;
}
-void create_rsync_argv(char ***argv, int64_t *num)
+static int handle_rm_exit(int status)
{
- char *logname, *newest = name_of_newest_complete_snapshot();
- int i = 0, j;
-
- *argv = dss_malloc((15 + conf.rsync_option_given) * sizeof(char *));
- (*argv)[i++] = dss_strdup("rsync");
- (*argv)[i++] = dss_strdup("-aq");
- (*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);
- } else
- DSS_INFO_LOG("no previous snapshot found\n");
- if (conf.exclude_patterns_given) {
- (*argv)[i++] = dss_strdup("--exclude-from");
- (*argv)[i++] = dss_strdup(conf.exclude_patterns_arg);
-
+ if (!WIFEXITED(status)) {
+ snapshot_removal_status = HS_READY;
+ return -E_INVOLUNTARY_EXIT;
}
- logname = dss_logname();
- if (conf.remote_user_given && !strcmp(conf.remote_user_arg, logname))
- (*argv)[i++] = dss_strdup(conf.source_dir_arg);
- else
- (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
- conf.remote_user_arg : logname,
- conf.remote_host_arg, conf.source_dir_arg);
- free(logname);
- *num = get_current_time();
- (*argv)[i++] = incomplete_name(*num);
- (*argv)[i++] = NULL;
- for (j = 0; j < i; j++)
- DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
+ if (WEXITSTATUS(status)) {
+ snapshot_removal_status = HS_READY;
+ return -E_BAD_EXIT_CODE;
+ }
+ snapshot_removal_status = HS_SUCCESS;
+ return 1;
}
-void free_rsync_argv(char **argv)
+static void handle_post_remove_exit(void)
{
- int i;
- for (i = 0; argv[i]; i++)
- free(argv[i]);
- free(argv);
+ snapshot_removal_status = HS_READY;
}
-int pre_create_hook(void)
+static int handle_remove_exit(int status)
{
- int ret, fds[3] = {0, 0, 0};
+ int ret;
+ struct snapshot *s = snapshot_currently_being_removed;
- if (!conf.pre_create_hook_given) {
- snapshot_creation_status = SCS_PRE_HOOK_SUCCESS;
- return 0;
+ 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;
}
- 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;
+ if (snapshot_removal_status == HS_READY) {
+ free(s->name);
+ free(s);
+ snapshot_currently_being_removed = NULL;
+ }
+ remove_pid = 0;
return ret;
}
-int post_create_hook(void)
+static int wait_for_remove_process(void)
{
- int ret, fds[3] = {0, 0, 0};
- char *cmd;
-
- if (!conf.post_create_hook_given) {
- snapshot_creation_status = SCS_READY;
- compute_next_snapshot_time();
- return 0;
- }
- cmd = make_message("%s %s", conf.post_create_hook_arg,
- path_to_last_complete_snapshot);
- DSS_NOTICE_LOG("executing %s\n", cmd);
- ret = dss_exec_cmdline_pid(&post_create_hook_pid, cmd, fds);
- free(cmd);
+ 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;
- snapshot_creation_status = SCS_POST_HOOK_RUNNING;
- return ret;
+ return handle_remove_exit(status);
}
-int create_snapshot(char **argv)
+static int handle_rsync_exit(int status)
{
- int ret, fds[3] = {0, 0, 0};
- char *name;
+ int es, 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);
+ if (!WIFEXITED(status)) {
+ 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
+ */
+ if (es == 12 || es == 13) {
+ DSS_WARNING_LOG("rsync process %d returned %d -- restarting\n",
+ (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)create_pid, es);
+ ret = -E_BAD_EXIT_CODE;
+ snapshot_creation_status = HS_READY;
+ goto out;
+ }
+ ret = rename_incomplete_snapshot(current_snapshot_creation_time);
if (ret < 0)
- return ret;
- snapshot_creation_status = SCS_RSYNC_RUNNING;
+ goto out;
+ snapshot_creation_status = HS_SUCCESS;
+ free(name_of_reference_snapshot);
+ name_of_reference_snapshot = NULL;
+out:
+ create_process_stopped = 0;
return ret;
}
-int handle_pre_create_hook_exit(int status)
+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;
}
-int handle_sigchld()
+static int handle_sigchld(void)
{
pid_t pid;
int status, ret = reap_child(&pid, &status);
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);
- exit(EXIT_FAILURE);
+ 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);
+ 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);
+ return -E_INVALID_NUMBER;
+ }
+ DSS_DEBUG_LOG("number of intervals: %i\n", conf.num_intervals_arg);
+ 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, config_file_exists;
+ char *config_file;
+ struct stat statbuf;
+ char *old_logfile_arg = NULL;
+ int old_daemon_given = 0;
+
+ if (conf.config_file_given)
+ config_file = dss_strdup(conf.config_file_arg);
+ else {
+ char *home = get_homedir();
+ config_file = make_message("%s/.dssrc", home);
+ free(home);
+ }
+ if (override) { /* SIGHUP */
+ if (conf.logfile_given)
+ old_logfile_arg = dss_strdup(conf.logfile_arg);
+ old_daemon_given = conf.daemon_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 (config_file_exists) {
+ struct cmdline_parser_params params = {
+ .override = override,
+ .initialize = 0,
+ .check_required = 1,
+ .check_ambiguity = 0,
+ .print_errors = 1
+ };
+ if (override) { /* invalidate all rsync options */
+ int i;
+
+ for (i = 0; i < conf.rsync_option_given; i++) {
+ free(conf.rsync_option_arg[i]);
+ conf.rsync_option_arg[i] = NULL;
+ }
+ conf.rsync_option_given = 0;
+ }
+ cmdline_parser_config_file(config_file, &conf, ¶ms);
+ }
+ ret = check_config();
+ if (ret < 0)
+ goto out;
+ if (override) {
+ /* don't change daemon mode on SIGHUP */
+ conf.daemon_given = old_daemon_given;
+ close_log(logfile);
+ logfile = NULL;
+ if (conf.logfile_given)
+ free(old_logfile_arg);
+ else if (conf.daemon_given) { /* re-use old logfile */
+ conf.logfile_arg = old_logfile_arg;
+ conf.logfile_given = 1;
+ }
+ }
+ 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);
+ ret = config_file_exists;
+out:
+ free(config_file);
+ if (ret < 0)
+ 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);
+ return dss_chdir(conf.dest_dir_arg);
+}
+
+static int handle_sighup(void)
+{
+ int ret;
+
+ 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();
}
-void handle_signal(void)
+static int handle_signal(void)
{
int sig, ret = next_signal();
switch (sig) {
case SIGINT:
case SIGTERM:
- restart_rsync_process();
- kill_process(rsync_pid);
- kill_process(rm_pid);
- exit(EXIT_FAILURE);
+ restart_create_process();
+ dss_kill(create_pid, SIGTERM, NULL);
+ dss_kill(remove_pid, SIGTERM, NULL);
+ ret = -E_SIGNAL;
+ break;
case SIGHUP:
- handle_sighup();
- ret = 1;
+ ret = handle_sighup();
break;
case SIGCHLD:
ret = handle_sigchld();
out:
if (ret < 0)
DSS_ERROR_LOG("%s\n", dss_strerror(-ret));
+ return ret;
}
-int get_oldest(const char *dirname, void *private)
+/*
+ * We can not use rsync locally if the local user is different from the remote
+ * user or if the src dir is not on the local host (or both).
+ */
+static int use_rsync_locally(char *logname)
{
- struct edge_snapshot_data *esd = private;
- struct snapshot s;
- int ret = is_snapshot(dirname, esd->now, &s);
+ char *h = conf.remote_host_arg;
- if (ret <= 0)
- return 1;
- if (s.creation_time > esd->snap.creation_time)
- return 1;
- free(esd->snap.name);
- esd->snap = s;
+ if (strcmp(h, "localhost") && strcmp(h, "127.0.0.1"))
+ return 0;
+ if (conf.remote_user_given && strcmp(conf.remote_user_arg, logname))
+ return 0;
return 1;
}
-int remove_oldest_snapshot()
+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;
- struct edge_snapshot_data esd = {
- .now = get_current_time(),
- .snap = {.creation_time = LLONG_MAX}
- };
- for_each_subdir(get_oldest, &esd);
- if (!esd.snap.name) /* no snapshot found */
- return 0;
- DSS_INFO_LOG("oldest snapshot: %s\n", esd.snap.name);
+ const char *why;
+
ret = 0;
- if (esd.snap.creation_time == current_snapshot_creation_time)
- goto out; /* do not remove the snapshot currently being created */
- ret = remove_snapshot(&esd.snap);
+ 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:
- free(esd.snap.name);
+ 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;
}
-/* TODO: Also consider number of inodes. */
-int disk_space_low(void)
+static void create_rsync_argv(char ***argv, int64_t *num)
{
- struct disk_space ds;
- int ret = get_disk_space(".", &ds);
+ char *logname;
+ int i = 0, j;
+ struct snapshot_list sl;
- 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;
- return 0;
+ dss_get_snapshot_list(&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("rsync");
+ (*argv)[i++] = dss_strdup("-aq");
+ (*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);
+ (*argv)[i++] = make_message("--link-dest=../%s",
+ name_of_reference_snapshot);
+ } else
+ 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);
+ else
+ (*argv)[i++] = make_message("%s@%s:%s/", conf.remote_user_given?
+ conf.remote_user_arg : logname,
+ conf.remote_host_arg, conf.source_dir_arg);
+ free(logname);
+ *num = get_current_time();
+ (*argv)[i++] = incomplete_name(*num);
+ (*argv)[i++] = NULL;
+ for (j = 0; j < i; j++)
+ DSS_DEBUG_LOG("argv[%d] = %s\n", j, (*argv)[j]);
+}
+
+static void free_rsync_argv(char **argv)
+{
+ int i;
+
+ if (!argv)
+ return;
+ for (i = 0; argv[i]; i++)
+ free(argv[i]);
+ free(argv);
}
-int try_to_free_disk_space(int low_disk_space)
+static int create_snapshot(char **argv)
{
int ret;
- struct snapshot_list sl;
- 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)
- goto out;
- DSS_WARNING_LOG("disk space low and nothing obvious to remove\n");
- ret = remove_oldest_snapshot();
- if (ret)
- goto out;
- DSS_CRIT_LOG("uhuhu: not enough disk space for a single snapshot\n");
- ret= -ENOSPC;
-out:
- free_snapshot_list(&sl);
+ ret = rename_resume_snap(current_snapshot_creation_time);
+ if (ret < 0)
+ return ret;
+ dss_exec(&create_pid, argv[0], argv);
+ snapshot_creation_status = HS_RUNNING;
return ret;
}
-int select_loop(void)
+static int select_loop(void)
{
int ret;
- struct timeval tv = {.tv_sec = 0, .tv_usec = 0};
+ /* check every 60 seconds for free disk space */
+ struct timeval tv;
+ char **rsync_argv = NULL;
for (;;) {
fd_set rfds;
- int low_disk_space;
- char **rsync_argv;
- struct timeval now, *tvp = &tv;
-
- if (rsync_pid)
- tv.tv_sec = 60; /* check every 60 seconds for free disk space */
- else if (rm_pid)
- tvp = NULL; /* sleep until rm process dies */
+ struct timeval *tvp;
+
+ if (remove_pid)
+ tvp = NULL; /* sleep until rm hook/process dies */
+ else { /* sleep one minute */
+ tv.tv_sec = 60;
+ tv.tv_usec = 0;
+ tvp = &tv;
+ }
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)
- return ret;
- if (FD_ISSET(signal_pipe, &rfds))
- handle_signal();
- if (rm_pid)
+ goto out;
+ if (FD_ISSET(signal_pipe, &rfds)) {
+ ret = handle_signal();
+ if (ret < 0)
+ goto out;
+ }
+ if (remove_pid)
continue;
- ret = disk_space_low();
- if (ret < 0)
- break;
- low_disk_space = ret;
- if (low_disk_space)
- stop_rsync_process();
- ret = try_to_free_disk_space(low_disk_space);
- if (ret < 0)
- break;
- if (rm_pid)
+ if (snapshot_removal_status == HS_PRE_SUCCESS) {
+ ret = exec_rm();
+ if (ret < 0)
+ goto out;
continue;
- restart_rsync_process();
- gettimeofday(&now, NULL);
- if (tv_diff(&next_snapshot_time, &now, &tv) > 0)
+ }
+ if (snapshot_removal_status == HS_SUCCESS) {
+ post_remove_hook();
+ continue;
+ }
+ ret = try_to_free_disk_space();
+ if (ret < 0)
+ goto out;
+ if (snapshot_removal_status != HS_READY) {
+ stop_create_process();
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_PRE_HOOK_SUCCESS:
- create_rsync_argv(&rsync_argv, ¤t_snapshot_creation_time);
+ 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);
- free_rsync_argv(rsync_argv);
if (ret < 0)
goto out;
continue;
- case SCS_RSYNC_RUNNING:
- continue;
- case SCS_RSYNC_SUCCESS:
- ret = post_create_hook();
+ case HS_NEEDS_RESTART:
+ if (!next_snapshot_is_due())
+ continue;
+ ret = create_snapshot(rsync_argv);
if (ret < 0)
goto out;
continue;
- case SCS_POST_HOOK_RUNNING:
+ case HS_SUCCESS:
+ post_create_hook();
continue;
}
}
return ret;
}
-int com_run(void)
+static void exit_hook(int exit_code)
+{
+ char *argv[3];
+ pid_t pid;
+
+ 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);
+}
+
+static int com_run(void)
{
int ret;
ret = install_sighandler(SIGHUP);
if (ret < 0)
return ret;
- compute_next_snapshot_time();
- return select_loop();
-}
-
-void log_disk_space(struct disk_space *ds)
-{
- DSS_INFO_LOG("free: %uM/%uM (%u%%), %u%% inodes unused\n",
- ds->free_mb, ds->total_mb, ds->percent_free,
- ds->percent_free_inodes);
+ ret = select_loop();
+ if (ret >= 0) /* impossible */
+ ret = -E_BUG;
+ exit_hook(ret);
+ return ret;
}
-int com_prune(void)
+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);
- for (;;) {
- 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 (;;) {
- 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;
}
-int com_create(void)
+static int com_create(void)
{
int ret, status;
char **rsync_argv;
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;
}
-int com_ls(void)
+static int com_ls(void)
{
int i;
struct snapshot_list sl;
struct snapshot *s;
- get_snapshot_list(&sl);
- FOR_EACH_SNAPSHOT(s, i, &sl)
- dss_msg("%u\t%s\n", s->interval, s->name);
+
+ dss_get_snapshot_list(&sl);
+ FOR_EACH_SNAPSHOT(s, i, &sl) {
+ int64_t d = 0;
+ 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;
}
-static void setup_signal_handling(void)
+static int setup_signal_handling(void)
{
int ret;
signal_pipe = signal_init(); /* always successful */
ret = install_sighandler(SIGINT);
if (ret < 0)
- goto err;
+ return ret;
ret = install_sighandler(SIGTERM);
if (ret < 0)
- goto err;
- ret = install_sighandler(SIGCHLD);
- if (ret < 0)
- goto err;
- return;
-err:
- DSS_EMERG_LOG("could not install signal handlers\n");
- exit(EXIT_FAILURE);
+ return ret;
+ 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;
.override = 0,
.initialize = 1,
.check_required = 0,
- .check_ambiguity = 0
+ .check_ambiguity = 0,
+ .print_errors = 1
};
cmdline_parser_ext(argc, argv, &conf, ¶ms); /* aborts on errors */
- parse_config_file(0);
-
+ 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();
- setup_signal_handling();
+ ret = change_to_dest_dir();
+ if (ret < 0)
+ goto out;
+ dump_dss_config("startup");
+ ret = setup_signal_handling();
+ if (ret < 0)
+ goto out;
ret = call_command_handler();
+out:
if (ret < 0)
DSS_EMERG_LOG("%s\n", dss_strerror(-ret));
exit(ret >= 0? EXIT_SUCCESS : EXIT_FAILURE);
projects / dss.git / blobdiff