merge cluster fixes from release_2_0 branch

[l2tpns.git] / Changes

diff --git a/Changes b/Changes
index 47dd20f..72e1d6f 100644 (file)
--- a/Changes
+++ b/Changes
@@ -1,6 +1,8 @@
-* Fri Apr 1 2005 Brendan O'Dea <bod@optusnet.com.au> 2.1.0
+* Thu May 26 2005 Brendan O'Dea <bod@optusnet.com.au> 2.1.0
 - Add IPv6 support from Jonathan McDowell.
-- Add CHAP support from Jordan Hrycaj (work in progress).
+- Add CHAP support from Jordan Hrycaj.
+- Add interim accounting support from Vladislav Bjelic.
+- Negotiate MRU, default 1458 to avoid fragmentation.
 - Sanity check that cluster_send_session is not called from a child
   process.
 - Throttle outgoing LASTSEEN packets to at most one per second for a
@@ -40,7 +42,41 @@
   (general error, try another LNS) when out of IP addresses, and 3
   (adminstrative) for everything else (suggestion from Chris Gates).
 - Only send RADIUS stop record in sessionshutdown when there's an ip address.
-- Reset .die on master takeover.
+- Reset .die on master takeover (so that dying sessions don't have to
+  hang around until the new master has the same uptime as the old one).
+- Update .last_packet in cluster_handle_bytes only when there have
+  been bytes received from the modem (dead sessions were having the
+  idle timeout reset by stray packets).
+- Use cli_error() for error messages and help.
+- Add a Cisco-Avpair with intercept details to RADIUS Start/Stop
+  records.
+- Don't use LOG() macro in initdata() until the config struct has been
+  allocated (uses config->debug).
+- Initialise log_stream to stderr to catch errors before the config file
+  is read.
+- Fix leak in session freelist when initial RADIUS session allocation
+  fails.
+- Make "show running-config" a privileged command (contains clear text
+  shared secrets).
+- Add sessionctl plugin to provide drop/kill via nsctl.
+- Add handling of "throttle=N" RADIUS attributes.
+- Fix RADIUS indexing (should have 16K entries with 64 sockets).
+- Cluster changes from Michael, intended to prevent a stray master
+  from trashing a cluster:
+  + Ignore heartbeats from peers claiming to be the master before the
+    timeout on the old master has expired.
+  + A master receiving a stray heartbeat sends a unicast HB back, which
+    should cause the rogue to die due to the tie-breaker code.
+  + Keep probing the master for late heartbeats.
+  + Drop BGP as soon as we become master with the minumum required peers.
+  + Any PING seen from a master forces an election (rather than just
+    where basetime is zero).
+  + A slave which receives a LASTSEEN message (presumably a restarted
+    master) sends back new message type, C_MASTER which indicates the
+    address of the current master.
+- New config option: cluster_master_min_adv which determines the minimum
+  number of up to date slaves required before the master will drop
+  routes.
 
 * Fri Dec 17 2004 Brendan O'Dea <bod@optusnet.com.au> 2.0.13
 - Better cluster master collision resolution: keep a counter of state