Convert firewall to nftables.

Debian Buster uses the nftables framework by default.

12 files changed, 283 insertions, 615 deletions

diff --git a/production b/production
index 0ce2b4e..ec655ab 100644
--- a/production
+++ b/production
@@ -82,3 +82,7 @@ benjamin
 # hostnames resolving to a dynamic IP
 [DynDNS:children]
 benjamin
+
+# need dhcp client
+[dhclient:children]
+benjamin
diff --git a/roles/common/files/etc/network/if-post-down.d/iptables b/roles/common/files/etc/network/if-post-down.d/iptables
deleted file mode 100755
index d27977d..0000000
--- a/roles/common/files/etc/network/if-post-down.d/iptables
+++ /dev/null
@@ -1,36 +0,0 @@
-#!/bin/sh
-
-# A post-down hook to flush ip tables and delete custom chains in the
-# loaded v4 and v6 rulesets.
-# Copyright © 2013 Guilhem Moulin <guilhem@fripost.org>
-#
-# This program is free software: you can redistribute it and/or modify
-# it under the terms of the GNU General Public License as published by
-# the Free Software Foundation, either version 3 of the License, or
-# (at your option) any later version.
-#
-# This program is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this program.  If not, see <http://www.gnu.org/licenses/>.
-
-set -ue
-PATH=/usr/sbin:/usr/bin:/sbin:/bin
-
-# Ignore the loopback interface; run the script for ifdown only.
-[ "$IFACE" != lo -a "$MODE" = stop ] || exit 0
-
-case "$ADDRFAM" in
-    inet)  ipts=/sbin/iptables-save;  ipt=/sbin/iptables;;
-    inet6) ipts=/sbin/ip6tables-save; ipt=/sbin/ip6tables;;
-    *)     exit 0
-esac
-
-$ipts | sed -nr 's/^\*//p' | \
-while read table; do
-    $ipt -t "$table" -F
-    $ipt -t "$table" -X
-done
diff --git a/roles/common/files/etc/network/if-pre-up.d/iptables b/roles/common/files/etc/network/if-pre-up.d/iptables
deleted file mode 100755
index 2b83cdc..0000000
--- a/roles/common/files/etc/network/if-pre-up.d/iptables
+++ /dev/null
@@ -1,47 +0,0 @@
-#!/bin/bash
-
-# A pre-up hook to auto-(re)load the iptables rulesets whenever the
-# network is brought up. If the action fails, an alert message is passed
-# to syslogd.
-# Copyright © 2013 Guilhem Moulin <guilhem@fripost.org>
-#
-# This program is free software: you can redistribute it and/or modify
-# it under the terms of the GNU General Public License as published by
-# the Free Software Foundation, either version 3 of the License, or
-# (at your option) any later version.
-#
-# This program is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this program.  If not, see <http://www.gnu.org/licenses/>.
-
-set -uo pipefail
-PATH=/usr/sbin:/usr/bin:/sbin:/bin
-
-# NOTE: syslog starts after networking during the boot process, messages
-# won't be logged at boot time.
-log="/usr/bin/logger -st firewall"
-
-# Ignore the loopback interface; run the script for ifup only.
-[ "$IFACE" != lo -a "$MODE" = start ] || exit 0
-
-# We support only IPv4 and IPv6.
-[ "$ADDRFAM" = inet -o "$ADDRFAM" = inet6 ] || exit 0
-
-$log -p user.info -- "Loading $ADDRFAM firewall on interface $IFACE."
-
-case "$ADDRFAM" in
-    inet) iptr=/sbin/iptables-restore;  rules=rules.v4;;
-    inet6)iptr=/sbin/ip6tables-restore; rules=rules.v6;;
-esac
-rules="/etc/iptables/$rules"
-
-$iptr < $rules 2>&1 | $log -p user.err
-rv=$?
-
-[ $rv -gt 0 ] && $log -p user.alert \
-    "WARN: Failed to load iptables rulesets; the machine may be unprotected!"
-exit $rv
diff --git a/roles/common/files/usr/local/sbin/update-firewall b/roles/common/files/usr/local/sbin/update-firewall
new file mode 100755
index 0000000..957bdc1
--- /dev/null
+++ b/roles/common/files/usr/local/sbin/update-firewall
@@ -0,0 +1,61 @@
+#!/bin/bash
+
+set -ue
+PATH=/usr/sbin:/usr/bin:/sbin:/bin
+export PATH
+
+NFTABLES="/etc/nftables.conf"
+
+script="$(mktemp --tmpdir=/dev/shm)"
+oldrules="$(mktemp --tmpdir=/dev/shm)"
+newrules="$(mktemp --tmpdir=/dev/shm)"
+netns=
+cleanup(){
+    rm -f -- "$script" "$oldrules" "$newrules"
+    [ -z "$netns" ] || ip netns del "$netns"
+}
+trap cleanup EXIT INT TERM
+
+echo "flush ruleset" >"$script" # should be included already, but...
+cat <"$NFTABLES" >>"$script"
+
+ip netns add "nft-dryrun"
+netns="nft-dryrun"
+
+# clear sets in the old rules before diff'ing with the new ones
+nft list ruleset -sn >"$oldrules"
+ip netns exec "$netns" nft -f - <"$oldrules"
+ip netns exec "$netns" nft flush set inet filter fail2ban
+ip netns exec "$netns" nft flush set inet filter fail2ban6
+ip netns exec "$netns" nft list ruleset -sn >"$oldrules"
+
+declare -a INTERFACES=()
+for iface in /sys/class/net/*; do
+    idx="$(< "$iface/ifindex")"
+    INTERFACES[idx]="${iface#/sys/class/net/}"
+done
+
+# create dummy interfaces so we can use iif/oif in the nft rules
+# (we preserve indices to preserve canonical set representation)
+for idx in "${!INTERFACES[@]}"; do
+    [ "${INTERFACES[idx]}" != "lo" ] || continue
+    ip netns exec "$netns" ip link add "${INTERFACES[idx]}" index "$idx" type dummy
+done
+
+ip netns exec "$netns" nft -f - <"$script"
+ip netns exec "$netns" nft list ruleset -sn >"$newrules"
+ip netns del "$netns"
+netns=
+
+if [ ! -t 0 ] || [ ! -t 1 ]; then
+    diff -q -- "$oldrules" "$newrules" && exit 0 || exit 1
+elif ! diff -u --color=auto --label=a/ruleset --label=b/ruleset \
+                -- "$oldrules" "$newrules" && nft -f - <"$script"; then
+    read -p "Ruleset applied. Revert? [Y/n] " -r -t10 r || r="y"
+    if [ "${r,,[a-z]}" != "n" ]; then 
+        echo "Reverting..."
+        echo "flush ruleset" >"$script"
+        cat <"$oldrules"    >>"$script"
+        nft -f - <"$script"
+    fi
+fi
diff --git a/roles/common/files/usr/local/sbin/update-firewall.sh b/roles/common/files/usr/local/sbin/update-firewall.sh
deleted file mode 100755
index 8ef3ab9..0000000
--- a/roles/common/files/usr/local/sbin/update-firewall.sh
+++ /dev/null
@@ -1,445 +0,0 @@
-#!/bin/bash
-
-# Create iptables (v4 and v6) rules.  Unless one of [-f] or [-c] is
-# given, or if the ruleset is unchanged, a confirmation is asked after
-# loading the new rulesets; if the user answers No or doesn't answer,
-# the old ruleset is restored.  If the user answer Yes (or if the flag
-# [-f] is given), the new ruleset is made persistent (requires a pre-up
-# hook) by moving it to /etc/iptables/rules.v[46].
-#
-# The [-c] flag switch to dry-run (check) mode.  The rulesets are not
-# applied, but merely checked against the existing ones.  The return
-# value is 0 iff. they do not differ.
-#
-# This firewall is only targeted towards end-servers, not gateways.  In
-# particular, there is no NAT'ing at the moment.
-#
-# Dependencies: netmask(1)
-#
-# Copyright © 2013 Guilhem Moulin <guilhem@fripost.org>
-#
-# This program is free software: you can redistribute it and/or modify
-# it under the terms of the GNU General Public License as published by
-# the Free Software Foundation, either version 3 of the License, or
-# (at your option) any later version.
-#
-# This program is distributed in the hope that it will be useful,
-# but WITHOUT ANY WARRANTY; without even the implied warranty of
-# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
-# GNU General Public License for more details.
-#
-# You should have received a copy of the GNU General Public License
-# along with this program.  If not, see <http://www.gnu.org/licenses/>.
-
-set -ue
-PATH=/usr/sbin:/usr/bin:/sbin:/bin
-timeout=10
-
-force=0
-check=0
-verbose=0
-addrfam=
-
-secproto=esp # must match /etc/ipsec.conf; ESP is the default (vs AH/IPComp)
-if [ -x /usr/sbin/ipsec ] && /usr/sbin/ipsec status >/dev/null; then
-    ipsec=y
-else
-    ipsec=n
-fi
-
-fail2ban_re='^(\[[0-9]+:[0-9]+\]\s+)?-A f2b-\S'
-IPsec_re=" -m policy --dir (in|out) --pol ipsec --reqid [0-9]+ --proto $secproto -j ACCEPT$"
-declare -A rss=() tables=()
-
-usage() {
-    cat >&2 <<- EOF
-		Usage: $0 [OPTIONS]
-
-		Options:
-		    -f force:   no confirmation asked
-		    -c check:   check (dry-run) mode
-		    -v verbose: see the difference between old and new ruleset
-		    -4 IPv4 only
-		    -6 IPv6 only
-	EOF
-    exit 1
-}
-
-log() {
-    logger -st firewall -p user.info -- "$@"
-}
-fatal() {
-    logger -st firewall -p user.err  -- "$@"
-    exit 1
-}
-
-iptables() {
-    # Fake iptables/ip6tables(8); use the more efficient
-    # iptables-restore(8) instead.
-    echo "$@" >>"$new";
-}
-commit() {
-    # End a table
-    echo COMMIT >>"$new"
-}
-inet46() {
-    case "$1" in
-        4) echo "$2";;
-        6) echo "$3";;
-    esac
-}
-ipt-chains() {
-    # Define new (tables and) chains.
-    while [ $# -gt 0 ]; do
-        case "$1" in
-            ?*:*) echo ":${1%:*} ${1##*:} [0:0]";;
-            ?*)   echo "*$1";;
-        esac
-        shift
-    done >>"$new"
-}
-
-ipt-trim() {
-    # Remove dynamic chain/rules from the input stream, as they are
-    # automatically included by third-party servers (such as strongSwan
-    # or fail2ban).  The output is ready to be made persistent.
-    grep -Ev -e '^:f2b-\S' \
-             -e "$IPsec_re" \
-             -e '-j f2b-\S+$' \
-             -e "$fail2ban_re"
-}
-
-ipt-diff() {
-    # Get the difference between two rulesets.
-    if [ $verbose -eq 1 ]; then
-        diff -u -I '^#' --color=auto "$@"
-    else
-        diff -q -I '^#' "$@" >/dev/null
-    fi
-}
-
-ipt-persist() {
-    # Make the current ruleset persistent.  (Requires a pre-up hook
-    # script to load the rules before the network is configured.)
-
-    log "Making ruleset persistent... "
-    [ -d /etc/iptables ] || mkdir /etc/iptables
-
-    local f rs table
-    for f in "${!tables[@]}"; do
-        ipts=$(inet46 $f iptables ip6tables)-save
-        rs=/etc/iptables/rules.v$f
-
-        for table in ${tables[$f]}; do
-            ip netns exec $netns $ipts -t $table
-        done | ipt-trim >"$rs"
-        chmod 0600 "$rs"
-    done
-}
-
-ipt-revert() {
-    [ $check -eq 0 ] || return
-    log "Reverting to old ruleset... "
-
-    local rs
-    for f in "${!rss[@]}"; do
-        $(inet46 $f iptables ip6tables)-restore -c <"${rss[$f]}"
-        rm -f "${rss[$f]}"
-    done
-    exit 1
-}
-
-run() {
-    # Build and apply the firewall for IPv4/6.
-    local f="$1"
-    local ipt=$(inet46 $f iptables ip6tables)
-    tables[$f]=filter
-
-    # The default interface associated with this address.
-    local if=$( /bin/ip -$f -o route show to default scope global \
-              | sed -nr '/^default via \S+ dev (\S+).*/ {s//\1/p;q}' )
-
-    # Store the old (current) ruleset
-    local old=$(mktemp --tmpdir current-rules.v$f.XXXXXX) \
-          new=$(mktemp --tmpdir new-rules.v$f.XXXXXX)
-    for table in ${tables[$f]}; do
-        $ipt-save -ct $table
-    done >"$old"
-    rss[$f]="$old"
-
-    local fail2ban=0
-    # XXX: As of Wheezy, fail2ban is IPv4 only.  See
-    #      https://github.com/fail2ban/fail2ban/issues/39 for the current
-    #      state of the art.
-    if [ "$f" = 4 ] && which fail2ban-server >/dev/null; then
-        fail2ban=1
-    fi
-
-    # The usual chains in filter, along with the desired default policies.
-    ipt-chains filter INPUT:DROP FORWARD:DROP OUTPUT:DROP
-
-    if [ ! "$if" ]; then
-        # If the interface is not configured, we stop here and DROP all
-        # packets by default.  Thanks to the pre-up hook this tight
-        # policy will be activated whenever the interface goes up.
-        commit
-        mv "$new" /etc/iptables/rules.v$f
-        return 0
-    fi
-
-    # Fail2ban-specific chains and traps
-    if [ $fail2ban -eq 1 ]; then
-        echo ":fail2ban - [0:0]"
-        # Don't remove existing rules & traps in the current rulest
-        grep    -- '^:f2b-\S'     "$old" || true
-        grep -E -- ' -j f2b-\S+$' "$old" || true
-        grep -E -- "$fail2ban_re" "$old" || true
-    fi >>"$new"
-
-    if [ "$f" = 4 -o "$f" = 6 ] && [ "$ipsec" = y ]; then
-        # IPsec tunnels come first (IPv4 only).
-        grep -E -- "$IPsec_re" "$old" >>"$new" || true
-
-        # Allow any IPsec $secproto protocol packets to be sent and received.
-        iptables -A INPUT  -i $if -p $secproto -j ACCEPT
-        iptables -A OUTPUT -o $if -p $secproto -j ACCEPT
-    fi
-
-
-    ########################################################################
-    # DROP all RFC1918 addresses, martian networks, multicasts, ...
-    # Credits to http://newartisans.com/2007/09/neat-tricks-with-iptables/
-    #            http://baldric.net/loose-iptables-firewall-for-servers/
-
-    local ip
-    if [ "$f" = 4 ] && [ "$ipsec" = y ]; then
-        # Private-use networks (RFC 1918) and link local (RFC 3927)
-        local MyIPsec="$( ip -4 -o route show table 220 dev $if | sed 's/\s.*//' )"
-        local MyNetwork="$( ip -4 -o address show dev $if scope global \
-                          | sed -nr "s/^[0-9]+:\s+$if\s+inet\s(\S+).*/\1/p" \
-                          | while read ip; do
-                              for ips in $MyIPsec; do
-                                [ "$ips" = "$(netmask -nc "$ip" "$ips" | sed 's/^ *//')" ] || echo "$ip"
-                              done
-                            done
-                          )"
-        [ "$MyNetwork" ] && \
-        for ip in 10.0.0.0/8 172.16.0.0/12 192.168.0.0/16 169.254.0.0/16; do
-            # Don't lock us out if we are behind a NAT ;-)
-            for myip in $MyNetwork; do
-                [ "$ip" = "$(netmask -nc "$ip" "$myip" | sed 's/^ *//')" ] || echo "$ip"
-            done | uniq | while read ip; do iptables -A INPUT -i $if -s "$ip" -j DROP; done
-        done
-
-        # Other martian packets: "This" network, multicast, broadcast (RFCs
-        # 1122, 3171 and 919).
-        for ip in 0.0.0.0/8 224.0.0.0/4 240.0.0.0/4 255.255.255.255/32; do
-            iptables -A INPUT -i $if -s "$ip" -j DROP
-            iptables -A INPUT -i $if -d "$ip" -j DROP
-        done
-
-    elif [ "$f" = 6 ]; then
-        # Martian IPv6 packets: ULA (RFC 4193) and site local addresses
-        # (RFC 3879).
-        for ip in fc00::/7 fec0::/10; do
-            iptables -A INPUT -i $if -s "$ip" -j DROP
-            iptables -A INPUT -i $if -d "$ip" -j DROP
-        done
-    fi
-
-    # DROP INVALID packets immediately.
-    iptables -A INPUT  -m state --state INVALID -j DROP
-    iptables -A OUTPUT -m state --state INVALID -j DROP
-
-    # DROP bogus TCP packets.
-    iptables -A INPUT -p tcp -m tcp --tcp-flags FIN,SYN FIN,SYN -j DROP
-    iptables -A INPUT -p tcp -m tcp --tcp-flags SYN,RST SYN,RST -j DROP
-    iptables -A INPUT -p tcp \! --syn -m state --state NEW      -j REJECT --reject-with tcp-reset
-
-    # Allow all input/output to/from the loopback interface.
-    local localhost=$(inet46 $f '127.0.0.1/8' '::1/128')
-    iptables -A INPUT  -i lo -s "$localhost" -d "$localhost" -j ACCEPT
-    iptables -A OUTPUT -o lo -s "$localhost" -d "$localhost" -j ACCEPT
-    if [ "$f" = 4 ] && [ "$ipsec" = y ]; then
-        # Allow local access to our virtual IP
-        ip -4 -o route show table 220 dev $if \
-        | sed -nr 's/.*\ssrc\s+([[:digit:].]{7,15})(\s.*)?$/\1/p' \
-        | while read ips; do
-            iptables -A INPUT  -i lo -s "$ips" -d "$ips" -j ACCEPT
-            iptables -A OUTPUT -o lo -s "$ips" -d "$ips" -j ACCEPT
-        done
-    fi
-
-    # Prepare fail2ban.  We make fail2ban insert its rules in a
-    # dedicated chain, so that it doesn't mess up the existing rules.
-    [ $fail2ban -eq 1 ] && iptables -A INPUT -i $if -j fail2ban
-
-    if [ "$f" = 4 ]; then
-        # Allow only ICMP of type 0, 3 and 8.  The rate-limiting is done
-        # directly by the kernel (net.ipv4.icmp_ratelimit and
-        # net.ipv4.icmp_ratemask runtime options).  See icmp(7).
-        local t
-        for t in  'echo-reply' 'destination-unreachable' 'echo-request'; do
-            iptables -A INPUT  -p icmp -m icmp --icmp-type $t -j ACCEPT
-            iptables -A OUTPUT -p icmp -m icmp --icmp-type $t -j ACCEPT
-        done
-    elif [ $f = 6 ]; then
-        iptables -A INPUT  -p icmpv6 -j ACCEPT
-        iptables -A OUTPUT -p icmpv6 -j ACCEPT
-    fi
-
-
-    ########################################################################
-    # ACCEPT new connections to the services we provide, or to those we want
-    # to connect to.
-
-    sed -re 's/#.*//; /^\s*$/d' -e "s/^(in|out|inout)$f?(\s.*)/\1\2/" \
-            /etc/iptables/services | \
-    grep -Ev '^(in|out|inout)\S\s' | \
-    while read dir proto dport sport; do
-        # We add two entries per config line: we need to accept the new
-        # connection, and latter the reply.
-        local stNew=NEW,ESTABLISHED,RELATED
-        local stEst=ESTABLISHED,RELATED
-
-        # In-Out means full-duplex
-        [[ "$dir" =~ ^inout ]] && stEst="$stNew"
-
-        local iptNew= iptEst= optsNew= optsEst=
-        case "$dport" in
-            *,*|*:*) optsNew="--match multiport --dports $dport"
-                     optsEst="--match multiport --sports $dport";;
-            ?*)      optsNew="--dport $dport"
-                     optsEst="--sport $dport";;
-        esac
-        case "$sport" in
-            *,*|*:*) optsNew+=" --match multiport --sports $sport"
-                     optsEst+=" --match multiport --dports $sport";;
-            ?*)      optsNew+=" --sport $sport"
-                     optsEst+=" --dport $sport";;
-        esac
-        case "$dir" in
-            in|inout) iptNew="-A INPUT  -i";  iptEst="-A OUTPUT -o";;
-            out)      iptNew="-A OUTPUT -o";  iptEst="-A INPUT  -i";;
-            *) fatal "Error: Unknown direction: '$dir'."
-        esac
-
-        iptables $iptNew $if -p $proto $optsNew -m state --state $stNew -j ACCEPT
-        iptables $iptEst $if -p $proto $optsEst -m state --state $stEst -j ACCEPT
-    done
-
-    iptables -A OUTPUT -o $if -p tcp -j REJECT --reject-with tcp-reset
-    iptables -A OUTPUT -o $if -p udp -j REJECT --reject-with port-unreach
-    if [ "$f" = "4" ]; then
-        iptables -A OUTPUT -o $if -p icmp -j REJECT --reject-with icmp-host-unreachable
-        iptables -A OUTPUT -o $if         -j REJECT --reject-with icmp-host-prohibited
-    else
-        iptables -A OUTPUT -o $if -j REJECT
-    fi
-
-    ########################################################################
-    commit
-
-
-    local rv1=0 rv2=0 persistent=/etc/iptables/rules.v$f
-    local oldz=$(mktemp --tmpdir current-rules.v$f.XXXXXX)
-
-    # Reset the counters.  They are not useful for comparing and/or
-    # storing persistent ruleset.  (We don't use sed -i because we want
-    # to restore the counters when reverting.)
-    sed -r -e '/^:/ s/\[[0-9]+:[0-9]+\]$/[0:0]/' \
-           -e 's/^\[[0-9]+:[0-9]+\]\s+//' \
-           "$old" >"$oldz"
-
-    ip netns exec $netns $ipt-restore <"$new" || ipt-revert
-
-    for table in ${tables[$f]}; do
-       ip netns exec $netns $ipt-save -t $table
-    done >"$new"
-
-    ipt-diff --label="a/$ipt-save" --label="b/$ipt-save" "$oldz" "$new" || rv1=$?
-
-    if ! [ -f "$persistent" ] && [ -x /etc/network/if-pre-up.d/iptables ]; then
-        rv2=1
-    else
-        ipt-trim <"$new" | ipt-diff --label="a/rules.v$f" --label="b/$ipt-save" "$persistent" - || rv2=$?
-    fi
-
-    local update="Please run '${0##*/}'."
-    if [ $check -eq 0 ]; then
-        uniq "$new" | $ipt-restore || ipt-revert
-    else
-        if [ $rv1 -ne 0 ]; then
-            log "WARN: The IPv$f firewall is not up to date! $update"
-        fi
-        if [ $rv2 -ne 0 ]; then
-            log "WARN: The current IPv$f firewall is not persistent! $update"
-        fi
-    fi
-
-    rm -f "$oldz" "$new"
-    return $(( $rv1 | $rv2 ))
-}
-
-
-# Parse options
-while [ $# -gt 0 ]; do
-    case "$1" in
-        -?*) for (( k=1; k<${#1}; k++ )); do
-                o="${1:$k:1}"
-                case "$o" in
-                    4|6) addrfam="$o";;
-                    c) check=1;;
-                    f) force=1;;
-                    v) verbose=1;;
-                    *) usage;;
-                esac
-            done
-        ;;
-        *) usage;;
-    esac
-    shift
-done
-
-# If we are going to apply the ruleset, we should either have a TTY, or
-# use -f.
-if ! tty -s && [ $force -eq 0 ] && [ $check -eq 0 ]; then
-    echo "Error: Not a TTY. Try with -f (at your own risks!)" >&2
-    exit 1
-fi
-
-# Create an alternative net namespace in which we apply the ruleset, so
-# we can easily get a normalized version we can compare latter.  See
-# http://bugzilla.netfilter.org/show_bug.cgi?id=790
-netns="ipt-firewall-test-$$"
-ip netns add $netns
-
-trap 'ip netns del $netns 2>/dev/null || true; ipt-revert' SIGINT
-trap 'ip netns del $netns; rm -f "${rss[@]}"'              EXIT
-
-rv=0
-for f in ${addrfam:=4 6}; do
-    run $f || rv=$(( $rv | $? ))
-done
-
-if [ $force -eq 1 ]; then
-    # At the user's own risks...
-    ipt-persist
-
-elif [ $check -eq 1 ] || [ $rv -eq 0 ]; then
-    # Nothing to do, we're all set.
-    exit $rv
-
-else
-    echo "Try now to establish NEW connections to the machine."
-
-    read -n1 -t$timeout \
-         -p "Are you sure you want to use the new ruleset? (y/N) " \
-         ret 2>&1 || { [ $? -gt 128 ] && echo -n "Timeout..."; }
-    case "${ret:-N}" in
-        [yY]*) echo; ipt-persist
-        ;;
-        *) echo; ipt-revert
-        ;;
-    esac
-fi
diff --git a/roles/common/tasks/firewall.yml b/roles/common/tasks/firewall.yml
index 133b631..fd1ad92 100644
--- a/roles/common/tasks/firewall.yml
+++ b/roles/common/tasks/firewall.yml
@@ -1,41 +1,27 @@
-- name: Install some packages required for the firewall
-  apt: pkg={{ packages }}
-  vars:
-    packages:
-    - iptables
-    - netmask
-    - bsdutils
+- name: Install nftables
+  apt: pkg=nftables
 
-- name: Create directory /etc/iptables
-  file: path=/etc/iptables
-        state=directory
-        owner=root group=root
-        mode=0755
-
-- name: Generate /etc/iptables/services
-  template: src=etc/iptables/services.j2
-            dest=/etc/iptables/services
-            owner=root group=root
-            mode=0600
-
-- name: Copy /usr/local/sbin/update-firewall.sh
-  copy: src=usr/local/sbin/update-firewall.sh
-        dest=/usr/local/sbin/update-firewall.sh
+- name: Copy /usr/local/sbin/update-firewall
+  copy: src=usr/local/sbin/update-firewall
+        dest=/usr/local/sbin/update-firewall
         owner=root group=staff
         mode=0755
 
-- name: Make the rulesets persistent
-  copy: src=etc/network/{{ item }}
-        dest=/etc/network/{{ item }}
-        owner=root group=root
-        mode=0755
-  with_items:
-    - if-pre-up.d/iptables
-    - if-post-down.d/iptables
+- name: Copy /etc/nftables.conf
+  template: src=etc/nftables.conf.j2
+            dest=/etc/nftables.conf
+            owner=root group=root
+            mode=0644
 
 - name: Ensure the firewall is up to date
-  command: /usr/local/sbin/update-firewall.sh -c
+  command: /usr/local/sbin/update-firewall -c
   register: rv
   # A non-zero return value will make ansible stop and show stderr. This
   # is what we want.
   changed_when: rv.rc
+
+- name: Enable nftables.service
+  service: name=nftables enabled=yes
+
+- name: Start nftables.service
+  service: name=nftables state=started
diff --git a/roles/common/tasks/main.yml b/roles/common/tasks/main.yml
index 7fa7b20..02a745c 100644
--- a/roles/common/tasks/main.yml
+++ b/roles/common/tasks/main.yml
@@ -12,6 +12,7 @@
   tags:
     - firewall
     - iptables
+    - nftables
 
 - import_tasks: stunnel.yml
   tags: stunnel
diff --git a/roles/common/tasks/sysctl.yml b/roles/common/tasks/sysctl.yml
index ffda544..3bf3b4f 100644
--- a/roles/common/tasks/sysctl.yml
+++ b/roles/common/tasks/sysctl.yml
@@ -18,7 +18,7 @@
     - { name: 'net.ipv4.icmp_ratemask',   value: 6425 }
     - { name: 'net.ipv4.icmp_ratelimit',  value: 1000 }
 
-      # Disable paquet forwarding between interfaces (we are not a router).
+      # Disable packet forwarding between interfaces (we are not a router).
     - { name: 'net.ipv4.ip_forward',          value: 0 }
     - { name: 'net.ipv6.conf.all.forwarding', value: 0 }
 
diff --git a/roles/common/templates/etc/ipsec.conf.j2 b/roles/common/templates/etc/ipsec.conf.j2
index 0ff9fbb..6b3840f 100644
--- a/roles/common/templates/etc/ipsec.conf.j2
+++ b/roles/common/templates/etc/ipsec.conf.j2
@@ -20,7 +20,7 @@ conn %default
     leftsubnet     = {{ ipsec[inventory_hostname_short] | ipv4 }}/32
     leftid         = {{ inventory_hostname }}
     leftsigkey     = {{ inventory_hostname_short }}.pem
-    leftfirewall   = yes
+    leftfirewall   = no
     lefthostaccess = yes
     rightauth      = pubkey
     auto           = route
diff --git a/roles/common/templates/etc/iptables/services.j2 b/roles/common/templates/etc/iptables/services.j2
deleted file mode 100644
index 6dd5aae..0000000
--- a/roles/common/templates/etc/iptables/services.j2
+++ /dev/null
@@ -1,48 +0,0 @@
-# {{ ansible_managed }}
-# Do NOT edit this file directly!
-#
-# direction              protocol     destination port            source port
-# (in|out|inout)[46]?    (tcp|udp|..) (port|port:port|port,port)  (port|port:port|port,port)
-
-{% if groups.all | length > 1 %}
-inout   udp     500      500                            # ISAKMP
-{% if groups.NATed | length > 0 %}
-inout4  udp     4500     4500                           # IPsec NAT Traversal
-{% endif %}
-{% endif %}
-
-out     tcp     80,443                                  # HTTP/HTTPS
-out     udp     53                                      # DNS
-out     tcp     53                                      # DNS
-out     udp     67                                      # DHCP
-out     tcp     22                                      # SSH
-out     udp     123      123                            # NTP
-
-in      tcp     {{ ansible_port|default('22') }}        # SSH
-{% if 'LDAP-provider' in group_names %}
-in      tcp     636                                     # LDAPS
-{% elif 'MX' in group_names or 'lists' in group_names or 'nextcloud' in group_names %}
-out     tcp     636                                     # LDAPS
-{% endif %}
-{% if 'MX' in group_names %}
-in      tcp     25                                      # SMTP
-{% endif %}
-{% if 'out' in group_names or 'MSA' in group_names %}
-out     tcp     25                                      # SMTP
-{% endif %}
-{% if 'IMAP' in group_names %}
-in      tcp     993                                     # IMAPS
-in      tcp     4190                                    # MANAGESIEVE
-out     tcp     2703                                    # Razor2
-{% endif %}
-{% if 'MSA' in group_names %}
-in      tcp     465                                     # SMTP-AUTH
-in      tcp     587                                     # SMTP-AUTH
-{% endif %}
-{% if 'webmail' in group_names or 'lists' in group_names or 'wiki' in group_names or 'nextcloud' in group_names %}
-in      tcp     80,443                                  # HTTP/HTTPS
-{% endif %}
-{% if 'LDAP-provider' in group_names %}
-out     tcp     11371                                   # HKP
-out     tcp     43                                      # WHOIS
-{% endif %}
diff --git a/roles/common/templates/etc/network/if-up.d/ipsec.j2 b/roles/common/templates/etc/network/if-up.d/ipsec.j2
index caa5129..9f183d3 100755
--- a/roles/common/templates/etc/network/if-up.d/ipsec.j2
+++ b/roles/common/templates/etc/network/if-up.d/ipsec.j2
@@ -25,10 +25,9 @@ PATH=/usr/sbin:/usr/bin:/sbin:/bin
 
 # Only the device with the default, globally-scoped route, is of
 # interest here.
-ip="$( ip -4 -o route show to default scope global \
-     | sed -nr '/^default via (\S+) dev (\S+).*/ {s//\2 \1/p;q}' )"
-[ "${ip% *}" = "$IFACE" ] || exit 0
-ip="${ip##* }"
+iface="$( ip -o route show to default scope global \
+        | sed -nr '/^default via \S+ dev (\S+).*/ {s//\1/p;q}' )"
+[ "$iface" = "$IFACE" ] || exit 0
 
 vip="{{ ipsec[inventory_hostname_short] }}"
 vsubnet="{{ ipsec_subnet }}"
@@ -39,9 +38,9 @@ case "$MODE" in
            # in the absence of xfrm lookup (i.e., when there is no
            # matching IPsec Security Association).
            ip route replace prohibit "$vsubnet" proto static || true
-           ip route replace table 220 to "$vsubnet" via "$ip" dev "$IFACE" proto static src "$vip" || true
+           ip route replace table 220 to "$vsubnet" dev "$IFACE" proto static src "$vip" || true
     ;;
-    stop)  ip route del table 220 to "$vsubnet" via "$ip" dev "$IFACE" proto static src "$vip" || true
+    stop)  ip route del table 220 to "$vsubnet" dev "$IFACE" proto static src "$vip" || true
            ip route del prohibit "$vsubnet" proto static || true
            ip address del "$vip/32" dev "$IFACE" scope global || true
 esac
diff --git a/roles/common/templates/etc/nftables.conf.j2 b/roles/common/templates/etc/nftables.conf.j2
new file mode 100755
index 0000000..1e1fde2
--- /dev/null
+++ b/roles/common/templates/etc/nftables.conf.j2
@@ -0,0 +1,193 @@
+#!/usr/sbin/nft -f
+
+define in-tcp-ports = {
+    {{ ansible_port|default(22) }}
+{% if 'MX' in group_names %}
+  , 25      # SMTP
+{% endif %}
+{% if 'LDAP-provider' in group_names %}
+  , 636     # ldaps
+{% endif %}
+{% if 'IMAP' in group_names %}
+  , 993     # imaps
+  , 4190    # ManageSieve
+{% endif %}
+{% if 'MSA' in group_names %}
+  , 587     # submission [RFC4409]
+  , 465     # submission over TLS [RFC8314]
+{% endif %}
+{% if 'webmail' in group_names or 'lists' in group_names or 'wiki' in group_names or 'nextcloud' in group_names %}
+  , 80      # HTTP
+  , 443     # HTTP over SSL/TLS
+{% endif %}
+}
+
+define out-tcp-ports = {
+    22
+  , 80      # HTTP
+  , 443     # HTTP over SSL/TLS
+{% if 'out' in group_names or 'MSA' in group_names %}
+  , 25      # SMTP
+{% endif %}
+{% if 'LDAP-provider' in group_names %}
+  , 11371   # OpenPGP HTTP Keyserver
+  , 43      # whois
+{% elif 'MX' in group_names or 'lists' in group_names or 'nextcloud' in group_names %}
+  , 636     # ldaps
+{% endif %}
+{% if 'IMAP' in group_names %}
+  , 2703    # Razor2
+{% endif %}
+}
+
+
+###############################################################################
+
+flush ruleset
+
+table inet filter {
+    # blackholes
+    set fail2ban  { type ipv4_addr; timeout 10m; }
+    set fail2ban6 { type ipv6_addr; timeout 10m; }
+
+    chain input {
+        type filter hook input priority 0
+        policy drop
+
+        iif lo accept
+
+        # XXX Bullseye: this is a hack for the lack of reqid matches in
+        # nftables: we mark the esp packet and accept after decapsulation
+        # https://serverfault.com/questions/971735/how-to-match-reqid-in-nftables
+        # https://blog.fraggod.net/2016/09/25/nftables-re-injected-ipsec-matching-without-xt_policy.html
+        define IPsec.mark = 0x220
+        meta l4proto esp mark set mark | $IPsec.mark accept
+        ip saddr 172.16.0.0/24 ip daddr 172.16.0.7 mark & $IPsec.mark == $IPsec.mark accept
+
+        # rate-limiting is done directly by the kernel (net.ipv4.icmp_{ratelimit,ratemask} runtime options)
+        icmp   type { echo-reply, echo-request, destination-unreachable } counter accept
+        icmpv6 type { echo-reply, echo-request, destination-unreachable,
+                      packet-too-big, time-exceeded, parameter-problem } counter accept
+
+        # accept neighbour discovery for autoconfiguration, RFC 4890 sec. 4.4.1
+        icmpv6 type { 133,134,135,136,141,142 } ip6 hoplimit 255 counter accept
+
+        jump martian
+        jump invalid
+
+        udp sport  123 udp dport  123 ct state     related,established accept
+{% if groups.all | length > 1 %}
+        udp sport  500 udp dport  500 ct state new,related,established accept
+{% if groups.NATed | length > 0 %}
+        udp sport 4500 udp dport 4500 ct state new,related,established accept
+{% endif %}
+{% endif %}
+
+        udp sport 53 ct state related,established accept
+        tcp sport 53 ct state related,established accept
+{% if 'dhclient' in group_names %}
+        udp sport 67 ct state related,established accept
+{% endif %}
+
+        meta l4proto tcp ip  saddr @fail2ban  counter drop
+        meta l4proto tcp ip6 saddr @fail2ban6 counter drop
+
+        tcp dport $in-tcp-ports  ct state related,established accept
+        tcp dport $in-tcp-ports  ct state new counter accept
+        tcp sport $out-tcp-ports ct state related,established accept
+    }
+
+    chain output {
+        type filter hook output priority 0
+        policy drop
+
+        oif lo accept
+
+        # XXX Bullseye: unlike for input we can't use marks here,
+        # because by the time we see a packet to 172.16.0.0/24 we don't
+        # know if it'll be encapsulated
+        meta l4proto esp accept
+        ip saddr 172.16.0.7 ip daddr 172.16.0.0/24 accept
+
+        meta l4proto { icmp, icmpv6 } accept
+
+        jump martian
+        jump invalid
+
+        udp sport  123 udp dport  123 ct state new,related,established accept
+        udp sport  500 udp dport  500 ct state new,related,established accept
+        udp sport 4500 udp dport 4500 ct state new,related,established accept
+
+        udp dport 53 ct state new,related,established accept
+        tcp dport 53 ct state new,related,established accept
+{% if 'dhclient' in group_names %}
+        udp dport 67 ct state new,related,established accept
+{% endif %}
+
+        tcp sport $in-tcp-ports  ct state related,established accept
+        tcp dport $out-tcp-ports ct state related,established accept
+        tcp dport $out-tcp-ports ct state new counter accept
+
+        meta l4proto tcp counter reject with tcp reset
+        meta l4proto udp counter reject
+        counter reject
+    }
+
+    chain martian {
+        # bogon filter (cf. RFC 6890 for non-global ip addresses)
+        define invalid-ip = {
+            0.0.0.0/8          # this host, on this network (RFC 1122 sec. 3.2.1.3)
+{% if not ansible_default_ipv4.address | ipaddr('10.0.0.0/8') %}
+          , 10.0.0.0/8         # private-use (RFC 1918)
+{% endif %}
+          , 100.64.0.0/10      # shared address space (RFC 6598)
+          , 127.0.0.0/8        # loopback (RFC 1122, sec. 3.2.1.3)
+          , 169.254.0.0/16     # link local (RFC 3927)
+{% if not ansible_default_ipv4.address | ipaddr('172.16.0.0/12') %}
+          , 172.16.0.0/12      # private-use (RFC 1918)
+{% endif %}
+          , 192.0.0.0/24       # IETF protocol assignments (RFC 6890 sec. 2.1)
+          , 192.0.2.0/24       # documentation (RFC 5737)
+{% if not ansible_default_ipv4.address | ipaddr('192.168.0.0/16') %}
+          , 192.168.0.0/16     # private-use (RFC 1918)
+{% endif %}
+          , 198.18.0.0/15      # benchmarking (RFC 2544)
+          , 198.51.100.0/24    # documentation (RFC 5737)
+          , 203.0.113.0/24     # documentation (RFC 5737)
+          , 240.0.0.0/4        # reserved (RFC 1112, sec. 4)
+          , 255.255.255.255/32 # limited broadcast (RFC 0919, section 7)
+        }
+
+        define invalid-ip6 = {
+            ::1/128         # loopback address (RFC 4291)
+          , ::/128          # unspecified (RFC 4291)
+          , ::ffff:0:0/96   # IPv4-mapped address (RFC 4291)
+          , 100::/64        # discard-only address block (RFC 6666)
+          , 2001::/23       # IETF protocol assignments (RFC 2928)
+          , 2001::/32       # TEREDO (RFC 4380)
+          , 2001:2::/48     # benchmarking (RFC 5180)
+          , 2001:db8::/32   # documentation (RFC 3849)
+          , 2001:10::/28    # ORCHID (RFC 4843)
+          , 2002::/16       # 6to4 (RFC 3056)
+          , fc00::/7        # unique-local (RFC 4193)
+          , fe80::/10       # linked-scoped unicast (RFC 4291)
+        }
+
+        ip saddr $invalid-ip counter drop
+        ip daddr $invalid-ip counter drop
+
+        ip6 saddr $invalid-ip6 counter drop
+        ip6 daddr $invalid-ip6 counter drop
+    }
+
+    chain invalid {
+        ct state invalid counter reject
+
+        # drop bogus TCP packets
+        tcp flags & (fin|syn|rst|psh|ack|urg) == 0x0         counter drop # null packets
+        tcp flags != syn ct state new                        counter drop # SYN-flood attacks
+        tcp flags & (fin|syn|rst|psh|ack|urg) == fin|psh|urg counter drop # XMAS packets
+        tcp flags & (fin|syn) == fin|syn                     counter drop # bogus
+        tcp flags & (syn|rst) == syn|rst                     counter drop # bogus
+    }
+}