From: Mark Wooding Date: Sun, 11 Mar 2012 23:58:09 +0000 (+0000) Subject: Overhaul address classification for link-local and non-unicast addresses. X-Git-Url: https://git.distorted.org.uk/~mdw/firewall/commitdiff_plain/44f9582788168b13a9163138f5e55bba889712e7 Overhaul address classification for link-local and non-unicast addresses. The previous attempts just weren't working. Intead, assign them their own classes, and work things using the forwarding masks. There's a minor wrinkle, that we must handle forwarded packets differently from inbound ones if they involve link-local addresses, but this is handled with a fixup in the mangle INPUT chain. The other significant change here is that the mangle table is now responsible for selecting packets with bogus destination addresses for rejection -- though it can't do the rejection itself because of a kernel restriction. --- diff --git a/classify.m4 b/classify.m4 index 606d90d..59ac3d9 100644 --- a/classify.m4 +++ b/classify.m4 @@ -65,6 +65,17 @@ m4_divert(40)m4_dnl clearchain mangle:in-classify mangle:in-default mangle:out-classify clearchain mangle:local-source +## An unpleasant hack. We can't reject packets from the mangle table, so +## we mark packets with a bad destination and then detect this in the +## filter table. +clearchain mangle:bad-destination-address +BAD_DEST=0xf6f377d2 +ip46tables -t mangle -A bad-destination-address -j MARK --set-mark $BAD_DEST +ip46tables -t mangle -A bad-destination-address -j ACCEPT +for i in $inchains; do + ip46tables -A $i -m mark --mark $BAD_DEST -g bad-destination-address +done + ## Packets over the loopback interface are automatically trusted. All manner ## of weird stuff happens on lo, and it's best not to second-guess it. run ip46tables -t mangle -A in-classify -i lo -j ACCEPT @@ -98,15 +109,6 @@ for addr in $host_6addrs; do -s $addr done -## It's not valid to have a multicast address as a packet source: multicast -## routing is done away from the source, so a multicast address would make -## this impossible to do. So discard these packets. Also discard class-E -## IPv4 addresses, since they aren't assigned. -run iptables -t mangle -A in-classify -g bad-source-address \ - -s 224.0.0.0/3 -run ip6tables -t mangle -A in-classify -g bad-source-address \ - -s ff00::/8 - m4_divert(41)m4_dnl ## Define the important networks. for pass in 1 2; do @@ -115,6 +117,56 @@ m4_divert(42)m4_dnl done m4_divert(46)m4_dnl +## Special IPv4 source addresses. Forbid broadcast and multicast sources. +## Mark the special zero address and link-local addresses as such. (This +## also matches class-E addresses, which are probably permanently invalid.) +for i in 0.0.0.0 169.254.0.0/16; do + run iptables -t mangle -A in-classify -g mark-from-link -s $i +done +run iptables -t mangle -A in-classify -g bad-source-address \ + -s 224.0.0.0/3 +run iptables -t mangle -A in-classify -g bad-source-address \ + -m addrtype --src-type BROADCAST \ + +## Special IPv6 addresses. Format multicast sources, and mark zero and +## link local addresses. +for i in :: fe80::/10; do + run ip6tables -t mangle -A in-classify -g mark-from-link -s $i +done +run ip6tables -t mangle -A in-classify -g bad-source-address \ + -s ff00::/8 + +## Special IPv4 destination addresses. The zero address is invalid; mark +## link-local and recognized broadcast addresses as link-local. We leave +## multicast for later. +for i in 0.0.0.0 240.0.0.0/4; do + run iptables -t mangle -A out-classify -g bad-destination-address -d $i +done +run iptables -t mangle -A out-classify -g mark-to-link -d 169.254.0.0/16 +run iptables -t mangle -A out-classify -g mark-to-link \ + -m addrtype --dst-type BROADCAST + +## Special IPv6 destination addressses. The zero address is again invalid; +## mark link local addresses. We do multicast later. +run ip6tables -t mangle -A out-classify -g bad-destination-address \ + -d :: +run ip6tables -t mangle -A out-classify -g mark-to-link -d fe80::/10 + +## Now deal with multicast. Link-local multicast is detected as being +## link-local, so that we can prevent it being forwarded correctly. +clearchain mangle:out-classify-mcast +run iptables -t mangle -A out-classify-mcast -g mark-to-link \ + -d 224.0.0.0/24 +for i in 0 1 2 3 4 5 6 7 8 9 a b c d e f; do + run ip6tables -t mangle -A out-classify-mcast -g mark-to-link \ + -d ff${i}2::/16 +done +run ip46tables -t mangle -A out-classify-mcast -g mark-to-mcast +run iptables -t mangle -A out-classify -g out-classify-mcast \ + -d 224.0.0.0/4 +run ip6tables -t mangle -A out-classify -g out-classify-mcast \ + -d ff00::/8 + ## Build the input classification chains. There's one chain `in-IFACE' for ## each local interface. This chain does a further dispatch on the source ## address to the appropriate `mark-from-CLASS' chain for the source network @@ -268,6 +320,14 @@ run ip46tables -t mangle -A out-classify -g mark-to-$net_class_default run ip46tables -t mangle -A PREROUTING -j in-classify run ip46tables -t mangle -A PREROUTING -j out-classify +## Incoming stuff to or from a link-local address is OK. +run ip46tables -t mangle -A INPUT \ + -m mark --mark $to_link/$MASK_TO \ + -j MARK --or-mark $fwd_link +run ip46tables -t mangle -A INPUT \ + -m mark --mark $from_link/$MASK_FROM \ + -j MARK --or-mark $fwd_link + ## Now it's safe to let stuff through. for i in PREROUTING INPUT FORWARD OUTPUT POSTROUTING; do run ip46tables -t mangle -P $i ACCEPT diff --git a/icmp.m4 b/icmp.m4 index 460c838..93c2973 100644 --- a/icmp.m4 +++ b/icmp.m4 @@ -33,21 +33,6 @@ for type in echo-request echo-reply; do run ip6tables -A check-icmp -p icmpv6 --icmpv6-type $type -j RETURN done -## Certainly don't allow ping to broadcast or multicast addresses. -case $forward in - 1) - run iptables -A FORWARD -g forbidden \ - -p icmp --icmp-type echo-request \ - -m addrtype --dst-type BROADCAST - run iptables -A FORWARD -g forbidden \ - -p icmp --icmp-type echo-request \ - -d 224.0.0.0/8 - run ip6tables -A FORWARD -g forbidden \ - -p icmpv6 --icmpv6-type echo-request \ - -d ff00::/16 - ;; -esac - m4_divert(58)m4_dnl ## Other ICMP is basically benign, we claim. run ip46tables -A check-icmp -j ACCEPT diff --git a/local.m4 b/local.m4 index 251cf59..d084db4 100644 --- a/local.m4 +++ b/local.m4 @@ -35,10 +35,12 @@ m4_divert(-1) ## Define the available network classes. m4_divert(42)m4_dnl -defnetclass untrusted untrusted trusted -defnetclass trusted untrusted trusted safe noloop -defnetclass safe trusted safe noloop -defnetclass noloop trusted safe +defnetclass untrusted untrusted trusted mcast +defnetclass trusted untrusted trusted safe noloop mcast +defnetclass safe trusted safe noloop mcast +defnetclass noloop trusted safe mcast +defnetclass link +defnetclass mcast m4_divert(-1) m4_divert(26)m4_dnl @@ -236,22 +238,6 @@ run iptables -A inbound -j ACCEPT \ -s 172.29.198.0/23 \ -p udp --source-port $port_bootpc --destination-port $port_bootps -## Incoming multicast on a network interface associated with a trusted -## network is OK, since it must have originated there (or been forwarded, but -## we don't do that yet). -seen=:-: -for net in $allnets; do - eval class=\$net_class_$net - case $class in trusted) ;; *) continue ;; esac - for iface in $(net_interfaces FWHOST $net); do - case "$seen" in *:$iface:*) continue ;; esac - seen=$seen$iface: - run iptables -A inbound -j ACCEPT \ - -s 0.0.0.0 -d 224.0.0.0/24 \ - -i $iface - done -done - ## Allow incoming ping. This is the only ICMP left. run ip46tables -A inbound -j ACCEPT -p icmp