ntp: refresh IP addresses periodically

Refresh NTP sources specified by hostname periodically (every 2 weeks
by default) to avoid long-running instances using a server which is no
longer intended for service, even if it is still responding correctly
and would not be replaced as unreachable, and help redistributing load
in large pools like pool.ntp.org. Only one source is refreshed at a time
to not interrupt clock updates if there are multiple selectable servers.

The refresh directive configures the interval. A value of 0 disables
the periodic refreshment.

Suggested-by: Ask Bjørn Hansen <ask@develooper.com>

conf.c

@@ -252,6 +252,9 @@ static char *leapsec_tz = NULL;
 /* Name of the user to which will be dropped root privileges. */
 static char *user;
 
+/* Address refresh interval */
+static int refresh = 1209600; /* 2 weeks */
+
 /* NTS server and client configuration */
 static char *nts_dump_dir = NULL;
 static char *nts_ntp_server = NULL;
@@ -702,6 +705,8 @@ CNF_ParseLine(const char *filename, int number, char *line)
                     &ntp_ratelimit_burst, &ntp_ratelimit_leak);
   } else if (!strcasecmp(command, "refclock")) {
     parse_refclock(p);
+  } else if (!strcasecmp(command, "refresh")) {
+    parse_int(p, &refresh);
   } else if (!strcasecmp(command, "reselectdist")) {
     parse_double(p, &reselect_distance);
   } else if (!strcasecmp(command, "rtcautotrim")) {
@@ -2533,6 +2538,14 @@ CNF_GetPtpPort(void)
 
 /* ================================================== */
 
+int
+CNF_GetRefresh(void)
+{
+  return refresh;
+}
+
+/* ================================================== */
+
 char *
 CNF_GetNtsDumpDir(void)
 {

conf.h

@@ -159,6 +159,8 @@ extern double CNF_GetHwTsTimeout(void);
 
 extern int CNF_GetPtpPort(void);
 
+extern int CNF_GetRefresh(void);
+
 extern char *CNF_GetNtsDumpDir(void);
 extern char *CNF_GetNtsNtpServer(void);
 extern int CNF_GetNtsServerCertAndKeyFiles(const char ***certs, const char ***keys);

doc/chrony.conf.adoc

@@ -884,6 +884,19 @@ This would disable the time checks until the clock is updated for the first
 time, assuming the first update corrects the clock and later checks can work
 with correct time.
 
+[[refresh]]*refresh* _interval_::
+This directive specifies the interval (in seconds) between refreshing IP
+addresses of NTP sources specified by hostname. If the hostname no longer
+resolves to the currently used address, it will be replaced with one of the new
+addresses to avoid using a server which is no longer intended for service, even
+if it is still responding correctly and would not be replaced as unreachable.
+Only one source is refreshed at a time. The default value is 1209600 (2 weeks)
+and the maximum value is 2^31-1 (68 years). A value of 0 disables the periodic
+refreshment.
++
+The <<chronyc.adoc#refresh,*refresh*>> command can be used to refresh all
+sources immediately.
+
 === Source selection
 
 [[authselectmode]]*authselectmode* _mode_::

ntp_sources.c

@@ -32,6 +32,7 @@
 #include "sysincl.h"
 
 #include "array.h"
+#include "conf.h"
 #include "ntp_sources.h"
 #include "ntp_core.h"
 #include "ntp_io.h"
@@ -64,6 +65,7 @@ typedef struct {
                                    received from the source yet */
   uint32_t conf_id;             /* Configuration ID, which can be shared with
                                    different sources in case of a pool */
+  double last_resolving;        /* Time of last name resolving (monotonic) */
 } SourceRecord;
 
 /* Hash table of SourceRecord, its size is a power of two and it's never
@@ -389,6 +391,7 @@ add_source(NTP_Remote_Address *remote_addr, char *name, NTP_Source_Type type,
       record->pool_id = pool_id;
       record->tentative = 1;
       record->conf_id = conf_id;
+      record->last_resolving = SCH_GetLastEventMonoTime();
 
       record_lock = 0;
 
@@ -985,9 +988,11 @@ resolve_source_replacement(SourceRecord *record, int refreshment)
 {
   struct UnresolvedSource *us;
 
-  DEBUG_LOG("trying to replace %s (%s)",
+  DEBUG_LOG("%s %s (%s)", refreshment ? "refreshing" : "trying to replace",
             UTI_IPToString(&record->remote_addr->ip_addr), record->name);
 
+  record->last_resolving = SCH_GetLastEventMonoTime();
+
   us = MallocNew(struct UnresolvedSource);
   us->name = Strdup(record->name);
   /* Ignore the order of addresses from the resolver to not get
@@ -1042,6 +1047,45 @@ NSR_HandleBadSource(IPAddr *address)
 
 /* ================================================== */
 
+static void
+maybe_refresh_source(void)
+{
+  static double last_refreshment = 0.0;
+  SourceRecord *record, *oldest_record;
+  int i, min_interval;
+  double now;
+
+  min_interval = CNF_GetRefresh();
+
+  now = SCH_GetLastEventMonoTime();
+  if (min_interval <= 0 || now < last_refreshment + min_interval)
+    return;
+
+  last_refreshment = now;
+
+  for (i = 0, oldest_record = NULL; i < ARR_GetSize(records); i++) {
+    record = get_record(i);
+    if (!record->remote_addr || UTI_IsStringIP(record->name))
+      continue;
+
+    if (!oldest_record || oldest_record->last_resolving > record->last_resolving)
+      oldest_record = record;
+  }
+
+  if (!oldest_record)
+    return;
+
+  /* Check if the name wasn't already resolved in the last interval */
+  if (now < oldest_record->last_resolving + min_interval) {
+    last_refreshment = oldest_record->last_resolving;
+    return;
+  }
+
+  resolve_source_replacement(oldest_record, 1);
+}
+
+/* ================================================== */
+
 void
 NSR_RefreshAddresses(void)
 {
@@ -1179,6 +1223,8 @@ NSR_ProcessRx(NTP_Remote_Address *remote_addr, NTP_Local_Address *local_addr,
           remove_pool_sources(record->pool_id, 1, 0);
       }
     }
+
+    maybe_refresh_source();
   } else {
     NCR_ProcessRxUnknown(remote_addr, local_addr, rx_ts, message, length);
   }

test/simulation/147-refresh

@@ -25,7 +25,35 @@ check_file_messages "20.*192.168.123.2" 15 17 measurements.log || test_fail
 check_file_messages "20.*192.168.123.[345]" 31 33 measurements.log || test_fail
 rm -f tmp/measurements.log
 if check_config_h 'FEAT_DEBUG 1'; then
+	check_log_messages "refreshing 192.168.123" 3 3 || test_fail
 	check_log_messages "resolved_name.*still fresh" 3 3 || test_fail
 fi
 
+limit=1100
+client_server_conf="
+server nodes-1-2.net1.clk maxpoll 6
+pool nodes-3-4-5.net1.clk maxpoll 6 maxsources 3"
+client_conf+="
+refresh 128"
+chronyc_conf=""
+
+run_test || test_fail
+check_chronyd_exit || test_fail
+check_source_selection || test_fail
+check_packet_interval || test_fail
+check_sync || test_fail
+
+check_file_messages "20.*192.168.123.1" 0 0 measurements.log || test_fail
+check_file_messages "20.*192.168.123.2" 16 18 measurements.log || test_fail
+check_file_messages "20.*192.168.123.[345]" 50 55 measurements.log || test_fail
+rm -f tmp/measurements.log
+if check_config_h 'FEAT_DEBUG 1'; then
+	check_log_messages "refreshing 192.168.123" 8 8 || test_fail
+	check_log_messages "resolved_name.*still fresh" 8 8 || test_fail
+	check_log_messages "refreshing 192.168.123.2" 2 2 || test_fail
+	check_log_messages "refreshing 192.168.123.3" 2 2 || test_fail
+	check_log_messages "refreshing 192.168.123.4" 2 2 || test_fail
+	check_log_messages "refreshing 192.168.123.5" 2 2 || test_fail
+fi
+
 test_pass