README-3.9-beta3.mrouted,v 1.1.2.1 1998/03/01 03:00:20 fenner Exp IP Multicast Extensions for BSD-Derived Unix Systems Multicast Routing Daemon Release 3.9-beta3 February 28, 1998 available from parcftp.xerox.com, file pub/net-research/ipmulti/beta-test/mrouted3.9-beta3.tar.Z binaries: pub/net-research/ipmulti/beta-test/mrouted3.9-beta3-sparc-sunos41x.tar.Z pub/net-research/ipmulti/beta-test/mrouted3.9-beta3-sparc-solaris2.tar.Z pub/net-research/ipmulti/beta-test/mrouted3.9-beta3-i386-freebsd22.tar.Z pub/net-research/ipmulti/beta-test/mrouted3.9-beta3-alpha-osf1.tar.Z pub/net-research/ipmulti/beta-test/mrouted3.9-beta3-sgi-irix6.tar.Z Note: The 3.9 release is mrouted-only, and will run on top of a 3.5 kernel. It is a drop-in replacement for mrouted 3.5, 3.6, 3.7 or 3.8. NOTE WELL: This is a beta-test release of mrouted. The basic functionality has been extensively tested in CAIRN and other testbeds, but it is expected to have bugs. Please report bugs to Bill Fenner . The 3.9-beta3 release fixes the following bugs: o There was a bug handling routing updates which caused random black holes. o There was a race condition in the timer handlers causing free'd memory to sometimes get touched. o "allow_nonpruners" wasn't allowed in the configuration file (and almost nobody noticed! - probably a good sign) o When a prune times out and the source has been active "recently", mrouted now waits for further traffic instead of triggering a new prune. o mrouted now ignores unreachable routes when making a routing decision (previously it would blackhole, now it can find a less-specific) The 3.9-beta3 release has the following new features: o A "blaster" keyword for mrouted.conf, to turn on handling of routers (mostly ciscos) which overwhelm the socket buffers by blasting the whole routing table at once. o A "notransit" keyword; routes learned on a "notransit" vif will not be readvertised onto another "notransit" vif. o The 500kbps default rate limit on tunnels has been removed. o An ICMP listener which logs ICMP errors which appear to be in response to tunnel packets that we sent. o A tunnel traffic encapsulator, which encapsulates control traffic inside the tunnel instead of unicasting it "beside" the tunnel. This is turned off by default; use "beside off" to turn it on. o A "force_leaf" flag to ignore any potential neighbors on a given interface. ========= 3.9-beta2 June 11, 1997 The 3.9-beta2 release fixes the following bugs: o There was a bug in 3.9-beta1's raw socket buffer processing that would cause an immediate lockup on startup on some systems. o RSRR would not clear out the group membership information if further notification of changes to this route entry was not possible. There is no need to upgrade to 3.9-beta2 if you are not experiencing one of the aforementioned bugs. ========= 3.9-beta1 June 6, 1997 The 3.9-beta1 release has the following known bugs: o The startup message doesn't print properly if you have too many interfaces. The 3.9-beta1 release fixes the following bugs: o mrouted did not properly keep track of subordinates, and would not time out subordinateness. This caused 2 major problems: 1. pruning did not happen when there were equal-cost paths to the same multi-access link 2. subordinateness which did not get cancelled by a non-poisoned route (e.g. in the face of route filtering) did not time out, causing traffic to continue to flow. o mrouted's IGMPv2 processing when it is not the querier now conforms to draft-ietf-idmr-igmp-v2-06.txt Thanks to Lorenzo VICISANO for finding a problem. o mrouted is much more careful about forgetting prunes; 3.8 would forget prunes whenever any route change ocurred. The 3.9-beta1 release has the following new features: o Longer prune lifetimes (2 hours) by default. Prune lifetimes may be configured per-vif, with the "prune_lifetime N" mrouted.conf configuration file entry (where N is in seconds). This helps to work around the black holes caused on restart when you have a Cisco upstream which does not handle genid's; if this is your situation the recommended value is 300. o mrouted's behavior of flooding new routes by default at startup in order to speed healing of paths during startup can be turned off per-vif or globally with the "noflood" configuration option. Turning this option off means you are likely to experience black holes for a minute or two when you restart a router. The default is to flood for a minute or two until mrouted is able to learn subordinate relationships. o mrouted now retransmits prunes by default on point-to-point links. prune retransmission can be turned on or off per vif via the "rexmit_prunes [on|off]" mrouted.conf command. Prune retransmission helps on lossy links, and also helps when a router has forgotten about a prune (e.g. if it is out of memory and needs to shed state, or due to a bug). o The new "passive" mode causes mrouted to not actively send probes looking for neighbors. This allows a dialup link to become quiescent if there is no DVMRP neighbor on the other end. Configuring "passive" on both ends of a link will cause it to never come up. o mrouted defaults to not peering with DVMRP routers that do not prune. Use the "allow_nonpruners" mrouted.conf option on a vif on which you want to allow such peerings. o mrouted now allows route filtering. mrouted.conf syntax: accept 13/8 accepts all routes matching 13/8 (e.g. accepts 13.2.116/22). If you want to accept only exactly 13/8, use accept 13/8 exact deny 10/8 64/2 130/8 exact 172/8 exact denies some common MBone martians Only "accept" or "deny" is allowed, no combinations. Add "bidir" to apply the filter to output too, otherwise it's input only. Expected usage: - Providers filter routes that customers send them - Martian removal - Topology modification (e.g. don't let the existence of private tunnel foo out into the world). o mrouted now malloc's the buffer it uses for SIOCGIFCONF, to allow for more interfaces. Thanks to Danny Mitzel o mrouted now ignores multiple entries for a single interface name (temporary hack until mrouted understands interface aliases) o mrouted's "-d" flag has been modified to accept the names of the systems which you would like to debug. packet, prunes, routes, peers, cache, timeout, interface, membership, traceroute, igmp o mrouted now times neighbors out fater, and fully detects and ignores routes from one-way peerings. o mrouted's route processing has been sped up, especially at startup. o mrouted uses the biggest SO_RCVBUF the operating system allows (up to 256Kbytes) o mrouted uses TOS 0xc0 ("Internet Control") for DVMRP messages. =========== Release 3.8 November 29, 1995 The 3.8 release fixes the following bugs: o mrouted would fail to forget prunes when a neighbor went away, thus potentially sending traffic down a tunnel after the tunnel endpoint has gone down. This was due to some research code making it into the "emergency" 3.7 release, sigh. o mrouted could send prunes with negative lifetimes. This causes slightly higher prune traffic but shouldn't be any major problem. =========== Release 3.7 November 28, 1995 The 3.7 release fixes the following bugs: o mrouted now ignores route reports that include bogus netmasks. There was a bug in 3.5 that would mangle default routes into tens of bogus routes; this should prevent that bug from killing the MBONE. This solution can cause route flaps and black holes until the 3.5's are gone or all of the 3.5's neighbors are 3.7 . o mrouted now ignores duplicate routes. Ciscos and the above 3.5 bug could cause two copies of the same route to appear in a single routing update; mrouted would insert two copies of the same route into its routing table and wreak all sorts of havoc. o mrouted now sends a group-specific query for both retransmissions of a g-s query; previous versions sent a general query the second time. o mrouted now loops back multicasted mtrace responses and group-specific membership queries o mrouted now performs deterministic tiebreaking between two neighbors on the same vif. o mrouted now only does duplicate suppression on traceroute requests, not all traceroute packets, so that a loop can be nicely detected via a duplicate router instead of just a timeout. o the buffer size that mrouted uses has been increased to allow more than 16 hops in mtrace messages. o mtrace's hop-by-hop termination is now more likely to be correct. o mrinfo now waits for the responses to its retransmitted queries. The 3.7 release has the following new features: o The configuration file can accept a hostname as the other end of a tunnel. There must be a single name->ip mapping for the given name, however, or mrouted will fail to start up. o mrinfo now sends requests to all interfaces of a multihomed host. o mtrace's passive mode has been implemented. o The first screen of mtrace statistics is shorter and more likely to fit on one screen. =========== Release 3.6 June 26, 1995 The 3.6 release fixes the following bugs: o mrouted would dump core when attempting to report no routes (i.e. upon startup, if you have no enabled phyint's) o mrouted would dump core if requested to traceroute a source for which it had no route o neighbor flags were not always properly updated on probe or report o mrouted would sometimes reply to a multicast traceroute on a disabled phyint; now it uses the first configured phyint to reply to traceroutes. o host routes (i.e. netmask 0xffffffff) works now; it was discarding IGMP from the host because it was coming from the "broadcast address" of the subnet. o send_igmp() now treats the failure to send an mtrace or a neighbor reply as informational, as opposed to warning. o mrouted would go into an infinite loop trying to respond to a traceroute for a source with a netmask of 0xffffffff. o vifs_with_neighbors was not being reset if the mrouted was restarted with SIGHUP o the default route was not being properly advertised to neighbors (although it was accepted if it was advertised to it) o ANSI-fication for those who it helps, still-K&R-ish for those it doesn't. o mtrace now attempts to trace three hops past a non-responding router, in the hopes that it does support traceroute but just couldn't respond (i.e. unicast didn't work and it can't source multicast because all its phyints are disabled). o mrinfo now times out even on a multicast router. =========== Release 3.5 May 8, 1995 The 3.5 release has the following new features: o The kernel and mrouted make sure that each is the correct version, to prevent problems with mismatched kernel/mrouted versions. A too-old mrouted will die with the error: can't enable DVMRP routing in kernel: Option not supported by protocol o mrouted can accept and propogate a default route (essential for heirarchical multicast routing) o Kernel route cache keeps source-specific routes instead of subnet routes, eliminating hashing and longest-match problems. (allows classless routing, longest-match and default routing) o Cached kernel routes only get deleted if no traffic is flowing, to facilitate multicast traceroute o mrouted has a new configuration file parser, which provides better error messages than before, and allows named boundaries (see man page) o added "netmask" to phyint configuration, at the suggestion of Anders Klemets o System V and FreeBSD compatibility from John Brezak o phyint's can have additional subnets configured, for people with multiple subnets on one physical network. mrouted.conf syntax is altnet 1.2.3.0, or altnet 1.2.3.0/24 if you need to specify a different netmask. There can be as many altnet statements as you need. o both mrouted and the kernel now support classless addresses. o the kernel supports PIM assert processing by notifying the router when a packet arrives on the wrong interface o the kernel keeps additional counters, and mrouted can be compiled to support SNMP and the Multicast MIB o the packet classifier in the kernel now uses the following udp port ranges: [0, 16384) - lowest priority, unclassified [16384, 32768) - highest priority, i.e. audio [32768, 49152) - medium priority, i.e. whiteboard [49152, 65536) - low priority, i.e. video A future release of a session directory will allocate ports in these ranges. o the configuration code has been modified to default tunnels' rate_limit parameters to 500kbps. This is easily modified with a rate_limit keyword in mrouted.conf, but should be a good default for the MBONE in general. o The tunnel sending code now caches a route for ip_output(), this should help performance on machines with lots of tunnels. o Dispatching for de-capsulating packets is now via protosw[], making reception of other raw protocols more efficient o Neighbor capabilities are discovered via a bitmask as opposed to version number. o Multicast traceroute code improved o mrouted can be compiled with Routing Support for Resource Reservation (RSRR), required for RSVP. The 3.5 release fixes the following bugs: o The IGMPv2 query timeout field was interpreted as being in units of 200ms as opposed to 100ms, thus the maximum timeout was set to twice the expected value. This is not fatal, as mrouted always queries twice in the expectation that a packet could get loss, but it does make it less robust in the face of packet loss. o IGMP could report membership in local-only groups (i.e. 224.0.0.X) o IGMP could get confused by hearing its own new membership reports, thus a router would never perform fast leave. o IGMP could reset timers for the wrong interface. o mrouted put a bogus value in the maximum timeout field of IGMPv2 query packets. o Non-querier mrouters would respond to IGMP leave messages o mrouted was not performing fast leave properly o If the last member goes away on a transit network, the upstream router would stop forwarding even if there are downstream members. o Kernel hash function improved o Eliminated possibility of panic(): timeout in cache maintenance o Reordered resource allocation when sending upcall to handle failure properly o some endian-ness bugs squashed in mrouted, probably more to go. o Multicast traceroute could send a reply on a disabled interface. This release consists of the following files: README-3.8.mrouted - this file mrouted/* - version 3.8 of mrouted, mrinfo, map-mbone and mtrace. ifconfig/* - Changes to ifconfig to show multicast interfaces netstat/* - Diffs to netstat ping/* - sources for ping which support multicasting mtest/* - utility for testing multicast group membership MROUTED 3.8 Mrouted 3.8 has two optional features: SNMP and RSRR. RSRR support is required for running RSVP; the SNMP code, with the help of the ISODE snmpd, implements both the Multicast MIB and the DVMRP MIB. RSRR ---- Routing Support for Resource Reservations (RSRR) was contributed by Daniel Zappala . To enable RSRR support, uncomment the three lines starting with RSRR near the top of the Makefile and "make clean; make". Or use the prebuilt binary, mrouted.rsrr . RSRR allows RSVP to query mrouted for its routing entry for a particular source-group pair. Using the routing entry and the IP_MULTICAST_VIF socket call, RSVP can forward distinct control messages out each outgoing interface. This version of mrouted supports RSRR messages using a Unix datagram socket. RSRR currently includes two pairs of query-reply messages. RSVP sends an Initial Query when it starts. Mrouted responds with an Initial Reply that includes the set of vifs it is using, flagging those that are administratively disabled. When RSVP needs the routing entry for a source-group pair, it sends a Route Query. Mrouted responds with a Route Reply that includes the incoming vif and outgoing vifs for the source-group pair. RSVP may request route change notification by setting the notification bit in the Route Query. If mrouted can provide route change notification for the source-group pair, it sets the notification bit in its Route Reply. When the routing entry for the source-group pair changes, mrouted sends an unsolicited Route Reply containing the new routing information. The initial release of mrouted 3.5 did not support route change notification and always returned a Route Reply with the notification bit cleared. This release of mrouted provides route change notification when possible. SNMP ---- SNMP support was contributed by David Thaler . To enable SNMP support, uncomment the six lines near the top of the Makefile below the description of SNMP support, or use the prebuilt binary, mrouted.snmp or mrouted.rsrr.snmp . To link the SNMP-capable mrouted, you need the CMU libraries. See http://nic.merit.edu/~mbone/ for a full mrouted-snmp distribution. Make sure to add the "sysName", "sysContact", "sysVersion" and "sysLocation" variables to your /etc/mrouted.conf if you want them to provide anything other than default values. Example: sysName "tibia" sysContact "Bill Fenner +1 415 812-4816" sysVersion "SunOS 4.1.3 and mrouted 3.8" sysLocation "MAXC room, PARC building 35" The SNMP version of mrouted has an additional command line flag: -P snmp_port Specifies a port for SNMP communication (default 161). This option should be used when another SNMP daemon already exists. The preferred alternate port in this case is port 9161. The mstat(8) program allows querying of statistics using SNMP.