[firewall] / functions.m4

### -*-sh-*-
###
### Utility functions for firewall scripts
###
### (c) 2008 Mark Wooding
###

###----- Licensing notice ---------------------------------------------------
###
### 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 2 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, write to the Free Software Foundation,
### Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.

m4_divert(20)m4_dnl
###--------------------------------------------------------------------------
### Utility functions.

## doit COMMAND ARGS...
##
## If debugging, print the COMMAND and ARGS.  If serious, execute them.
run () {
  set -e
  if [ "$FW_DEBUG" ]; then echo "* $*"; fi
  if ! [ "$FW_NOACT" ]; then "$@"; fi
}

## trace MESSAGE...
##
## If debugging, print the MESSAGE.
trace () {
  set -e
  if [ "$FW_DEBUG" ]; then echo "$*"; fi
}

## defport NAME NUMBER
##
## Define $port_NAME to be NUMBER.
defport () {
  name=$1 number=$2
  eval port_$name=$number
}

## defproto NAME NUMBER
##
## Define $proto_NAME to be NUMBER.
defproto () {
  name=$1 number=$2
  eval proto_$name=$number
}

## addword VAR WORD
##
## Adds WORD to the value of the shell variable VAR, if it's not there
## already.  Words are separated by a single space; no leading or trailing
## spaces are introduced.
addword () {
  var=$1 word=$2
  eval val=\$$var
  case " $val " in
    *" $word "*) ;;
    *) eval "$var=\${$var:+\$val }\$word" ;;
  esac
}

m4_divert(38)m4_dnl
###--------------------------------------------------------------------------
### Utility chains (used by function definitions).

m4_divert(20)m4_dnl
###--------------------------------------------------------------------------
### Basic chain constructions.

## ip46tables ARGS ...
##
## Do the same thing for `iptables' and `ip6tables'.
ip46tables () {
  set -e
  iptables "$@"
  ip6tables "$@"
}

## clearchain CHAIN CHAIN ...
##
## Ensure that the named chains exist and are empty.
clearchain () {
  set -e
  for _chain; do
    case $_chain in
      *:*) table=${_chain%:*} _chain=${_chain#*:} ;;
      *) table=filter ;;
    esac
    run ip46tables -t $table -N $_chain 2>/dev/null || :
  done
}

## makeset SET TYPE [PARAMS]
##
## Ensure that the named ipset exists.  Don't clear it.
makeset () {
  set -e
  name=$1; shift
  v=$(ipset --version)
  createp=t
  case "$v" in
    "ipset v4"*)
      if ipset -nL | grep -q "^Name: $name\$"; then createp=nil; fi
      ;;
    *)
      if ipset -n -L | grep -q "^$name\$"; then createp=nil; fi
      ;;
  esac
  case $createp in
    t) ipset -N "$name" "$@" ;;
  esac
}

## errorchain CHAIN ACTION ARGS ...
##
## Make a chain which logs a message and then invokes some other action,
## typically REJECT.  Log messages are prefixed by `fw: CHAIN'.
errorchain () {
  set -e
  chain=$1; shift
  case $chain in
    *:*) table=${chain%:*} chain=${chain#*:} ;;
    *) table=filter ;;
  esac
  clearchain $table:$chain
  run ip46tables -t $table -A $chain -j LOG \
	  -m limit --limit 3/minute --limit-burst 10 \
	  --log-prefix "fw: $chain " --log-level notice || :
  run ip46tables -t $table -A $chain -j "$@" \
	  -m limit --limit 20/second --limit-burst 100
  run ip46tables -t $table -A $chain -j DROP
}

m4_divert(20)m4_dnl
###--------------------------------------------------------------------------
### Basic option setting.

## setopt OPTION VALUE
##
## Set an IP sysctl.
setopt () {
  set -e
  opt=$1 val=$2
  any=nil
  for ver in ipv4 ipv6; do
    if [ -f /proc/sys/net/$ver/$opt ]; then
      run sysctl -q net/$ver/$opt="$val"
      any=t
    fi
  done
  case $any in
    nil) echo >&2 "$0: unknown IP option $opt"; exit 1 ;;
  esac
}

## setdevopt OPTION VALUE [INTERFACES ...]
##
## Set an IP interface-level sysctl.
setdevopt () {
  set -e
  opt=$1 val=$2; shift 2
  case "$#,$1" in
    0, | 1,all)
      set -- $(
	seen=:
	for ver in ipv4 ipv6; do
	  cd /proc/sys/net/$ver/conf
	  for i in *; do
	    [ -f $i/$opt ] || continue
	    case "$seen" in (*:$i:*) continue ;; esac
	    echo $i
	  done
	done)
      ;;
  esac
  for i in "$@"; do
    any=nil
    for ver in ipv4 ipv6; do
      if [ -f /proc/sys/net/$ver/conf/$i/$opt ]; then
	any=t
	run sysctl -q net/$ver/conf/$i/$opt="$val"
      fi
    done
    case $any in
      nil) echo >&2 "$0: unknown device option $opt"; exit 1 ;;
    esac
  done
}

m4_divert(20)m4_dnl
###--------------------------------------------------------------------------
### Packet filter construction.

## conntrack CHAIN
##
## Add connection tracking to CHAIN, and allow obvious stuff.
conntrack () {
  set -e
  chain=$1
  run ip46tables -A $chain -p tcp -m state \
	  --state ESTABLISHED,RELATED -j ACCEPT
  run ip46tables -A $chain -p tcp ! --syn -g bad-tcp
}

## commonrules CHAIN
##
## Add standard IP filtering rules to the CHAIN.
commonrules () {
  set -e
  chain=$1

  ## Pass fragments through, assuming that the eventual destination will sort
  ## things out properly.  Except for TCP, that is, which should never be
  ## fragmented.  This is an extra pain for ip6tables, which doesn't provide
  ## a pleasant way to detect non-initial fragments.
  run iptables -A $chain -p tcp -f -g tcp-fragment
  run iptables -A $chain -f -j ACCEPT
  run ip6tables -A $chain -p tcp -g tcp-fragment \
	  -m ipv6header --soft --header frag
  run ip6tables -A $chain -j accept-non-init-frag
}

m4_divert(38)m4_dnl
## Accept a non-initial fragment.  This is only needed by IPv6, to work
## around a deficiency in the option parser.
run ip6tables -N accept-non-init-frag
run ip6tables -A accept-non-init-frag -j RETURN \
	-m frag --fragfirst
run ip6tables -A accept-non-init-frag -j ACCEPT

m4_divert(20)m4_dnl
## allowservices CHAIN PROTO SERVICE ...
##
## Add rules to allow the SERVICES on the CHAIN.
allowservices () {
  set -e
  chain=$1 proto=$2; shift 2
  count=0
  list=
  for svc; do
    case $svc in
      *:*)
	n=2
	left=${svc%:*} right=${svc#*:}
	case $left in *[!0-9]*) eval left=\$port_$left ;; esac
	case $right in *[!0-9]*) eval right=\$port_$right ;; esac
	svc=$left:$right
	;;
      *)
	n=1
	case $svc in *[!0-9]*) eval svc=\$port_$svc ;; esac
	;;
    esac
    case $svc in
      *: | :* | "" | *[!0-9:]*)
	echo >&2 "Bad service name"
	exit 1
	;;
    esac
    count=$(( $count + $n ))
    if [ $count -gt 15 ]; then
      run ip46tables -A $chain -p $proto -m multiport -j ACCEPT \
	     --destination-ports ${list#,}
      list= count=$n
    fi
    list=$list,$svc
  done
  case $list in
    "")
      ;;
    ,*,*)
      run ip46tables -A $chain -p $proto -m multiport -j ACCEPT \
	      --destination-ports ${list#,}
      ;;
    *)
      run ip46tables -A $chain -p $proto -j ACCEPT \
	      --destination-port ${list#,}
      ;;
  esac
}

## ntpclient CHAIN NTPSERVER ...
##
## Add rules to CHAIN to allow NTP with NTPSERVERs.
ntpclient () {
  set -e
  ntpchain=$1; shift

  clearchain ntp-servers
  for ntp; do run iptables -A ntp-servers -j ACCEPT -s $ntp; done
  run iptables -A $ntpchain -j ntp-servers \
	  -p udp --source-port 123 --destination-port 123
}

## dnsresolver CHAIN
##
## Add rules to allow CHAIN to be a DNS resolver.
dnsresolver () {
  set -e
  chain=$1
  for p in tcp udp; do
    run ip46tables -A $chain -j ACCEPT \
	     -m state --state ESTABLISHED \
	     -p $p --source-port 53
  done
}

## dnsserver CHAIN
##
## Add rules to allow CHAIN to be a DNS server.
dnsserver () {
  set -e
  chain=$1

  ## Allow TCP access.  Hitting us with SYNs will make us deploy SYN cookies,
  ## but that's tolerable.
  run ip46tables -A $chain -j ACCEPT -p tcp --destination-port 53

  ## Avoid being a DDoS amplifier by rate-limiting incoming DNS queries.
  clearchain $chain-udp-dns
  run ip46tables -A $chain-udp-dns -j ACCEPT \
	  -m limit --limit 20/second --limit-burst 300
  run ip46tables -A $chain-udp-dns -g dns-rate-limit
  run ip46tables -A $chain -j $chain-udp-dns \
	  -p udp --destination-port 53
}

## openports CHAIN [MIN MAX]
##
## Add rules to CHAIN to allow the open ports.
openports () {
  set -e
  chain=$1; shift
  [ $# -eq 0 ] && set -- $open_port_min $open_port_max
  run ip46tables -A $chain -p tcp -g interesting --destination-port $1:$2
  run ip46tables -A $chain -p udp -g interesting --destination-port $1:$2
}

m4_divert(20)m4_dnl
###--------------------------------------------------------------------------
### Packet classification.
###
### See `classify.m4' for an explanation of how the firewall machinery for
### packet classification works.
###
### A list of all network names is kept in `allnets'.  For each network NET,
### shell variables are defined describing their properties.
###
### net_class_NET	The class of the network, as defined by
###			`defnetclass'.
### net_inet_NET	List of IPv4 address ranges in the network.
### net_inet6_NET	List of IPv6 address ranges in the network.
### net_via_NET		List of other networks that this one forwards via.
### net_hosts_NET	List of hosts known to be in the network.
### host_inet_HOST	IPv4 address of the named HOST.
### host_inet6_HOST	IPv6 address of the named HOST.
###
### Similarly, a list of hosts is kept in `allhosts', and for each host HOST,
### a shell variables are defined:
###
### host_ifaces_HOST	List of interfaces for this host and the networks
###			they attach to, in the form IFACE=NET.

## defbitfield NAME WIDTH
##
## Defines MASK_NAME and BIT_NAME symbolic constants for dealing with
## bitfields: x << BIT_NAME yields the value x in the correct position, and
## ff & MASK_NAME extracts the corresponding value.
defbitfield () {
  set -e
  name=$1 width=$2
  eval MASK_$name=$(( (1 << $width) - 1 << $bitindex ))
  eval BIT_$name=$bitindex
  bitindex=$(( $bitindex + $width ))
}

## Define the layout of the bitfield.
bitindex=0
defbitfield MASK 16
defbitfield FROM 4
defbitfield TO 4

## defnetclass NAME FORWARD-TO...
##
## Defines a netclass called NAME, which is allowed to forward to the
## FORWARD-TO netclasses.
##
## For each netclass, constants from_NAME and to_NAME are defined as the
## appropriate values in the FROM and TO fields (i.e., not including any mask
## bits).
##
## This function also establishes mangle chains mark-from-NAME and
## mark-to-NAME for applying the appropriate mark bits to the packet.
##
## Because it needs to resolve forward references, netclasses must be defined
## in a two-pass manner, using a loop of the form
##
##   for pass in 1 2; do netclassindex=0; ...; done
netclassess=
defnetclass () {
  set -e
  name=$1; shift
  case $pass in
    1)

      ## Pass 1.  Establish the from_NAME and to_NAME constants, and the
      ## netclass's mask bit.
      trace "netclass $name = $netclassindex"
      eval from_$name=$(( $netclassindex << $BIT_FROM ))
      eval to_$name=$(( $netclassindex << $BIT_TO ))
      eval fwd_$name=$(( 1 << ($netclassindex + $BIT_MASK) ))
      nets="$nets $name"
      ;;
    2)

      ## Pass 2.  Compute the actual from and to values.  This is fiddly:
      ## we want to preserve the other flags.
      from=$(( ($netclassindex << $BIT_FROM) ))
      frommask=$(( $MASK_FROM | $MASK_MASK ))
      for net; do
	eval bit=\$fwd_$net
	from=$(( $from + $bit ))
      done
      to=$(( ($netclassindex << $BIT_TO) ))
      tomask=$(( $MASK_TO | $MASK_MASK ^ (1 << ($netclassindex + $BIT_MASK)) ))
      trace "from $name --> set $(printf %08x/%08x $from $frommask)"
      trace "  to $name --> set $(printf %08x/%08x $to $tomask)"

      ## Now establish the mark-from-NAME and mark-to-NAME chains.
      clearchain mangle:mark-from-$name mangle:mark-to-$name
      run ip46tables -t mangle -A mark-from-$name -j MARK \
	      --set-xmark $from/$frommask
      run ip46tables -t mangle -A mark-to-$name -j MARK \
	      --set-xmark $to/$tomask
      ;;
  esac
  netclassindex=$(( $netclassindex + 1 ))
}

## defnet NET CLASS
##
## Define a network.  Follow by calls to `addr', `via', etc. to define
## properties of the network.  Networks are processed in order, so if their
## addresses overlap then the more specific addresses should be defined
## earlier.
defnet () {
  net=$1 class=$2
  addword allnets $net
  eval net_class_$1=\$class
}

## addr ADDRESS/LEN ...
##
## Define addresses for the network being defined.  ADDRESSes are in
## colon-separated IPv6 or dotted-quad IPv4 form.
addr () {
  for i in "$@"; do
    case "$i" in
      *:*) addword net_inet6_$net $i ;;
      *) addword net_inet_$net $i ;;
    esac
  done
}

## via NET ...
##
## Declare that packets from this network are forwarded to the other NETs.
via () {
  eval "net_via_$net=\"$*\""
}

## noxit NET ...
##
## Declare that packets from this network must not be forwarded to the other
## NETs.
noxit () {
  eval "net_noxit_$net=\"$*\""
}

## host HOST ADDR ...
##
## Define the address of an individual host on the current network.  The
## ADDRs may be full IPv4 or IPv6 addresses, or offsets from the containing
## network address, which is a simple number for IPv4, or a suffix beginning
## with `::' for IPv6.  If an IPv6 base address is provided for the network
## but not for the host then the host's IPv4 address is used as a suffix.
host () {
  name=$1; shift

  ## Work out which addresses we've actually been given.
  unset a6
  for i in "$@"; do
    case "$i" in ::*) a6=$i ;; *) a=$i ;; esac
  done
  case "${a+t}" in
    t) ;;
    *) echo >&2 "$0: no address for $name"; exit 1 ;;
  esac
  case "${a6+t}" in t) ;; *) a6=::$a ;; esac

  ## Work out the IPv4 address.
  eval nn=\$net_inet_$net
  for n in $nn; do
    addr=${n%/*}
    base=${addr%.*}
    offset=${addr##*.}
    case $a in *.*) aa=$a ;; *) aa=$base.$(( $offset + $a )) ;; esac
    eval host_inet_$name=$aa
  done

  ## Work out the IPv6 address.
  eval nn=\$net_inet6_$net
  for n in $nn; do
    addr=${n%/*}
    base=${addr%::*}
    case $a6 in ::*) aa=$base$a6 ;; *) aa=$a6 ;; esac
    eval host_inet6_$name=$aa
  done

  ## Remember the host in the list.
  addword net_hosts_$net $name
}

## defhost NAME
##
## Define a new host.  Follow by calls to `iface' to define the host's
## interfaces.
defhost () {
  host=$1
  addword allhosts $host
  eval host_type_$host=server
}

## hosttype TYPE
##
## Declare the host to have the given type.
hosttype () {
  type=$1
  case $type in
    router | server | client) ;;
     *) echo >&2 "$0: bad host type \`$type'"; exit 1 ;;
  esac
  eval host_type_$host=$type
}

## iface IFACE NET ...
##
## Define a host's interfaces.  Specifically, declares that the host has an
## interface IFACE attached to the listed NETs.
iface () {
  name=$1; shift
  for net in "$@"; do
    addword host_ifaces_$host $name=$net
  done
}

## matchnets OPT WIN FLAGS PREPARE BASE SUFFIX NEXT NET [NET ...]
##
## Build rules which match a particular collection of networks.
##
## Specifically, use the address-comparison operator OPT (typically `-s' or
## `-d') to match the addresses of each NET, writing the rules to the chain
## BASESUFFIX.  If we find a match, dispatch to WIN-CLASS, where CLASS is the
## class of the matching network.  In order to deal with networks containing
## negative address ranges, more chains may need to be constructed; they will
## be named BASE#Q for sequence numbers Q starting with NEXT.  All of this
## happens on the `mangle' table, and there isn't (currently) a way to tweak
## this.
##
## The FLAGS gather additional interesting information about the job,
## separated by colons.  The only flag currently is :default: which means
## that the default network was listed.
##
## Finally, there is a hook PREPARE which is called just in advance of
## processing the final network, passing it the argument FLAGS.  (The PREPARE
## string will be subjected to shell word-splitting, so it can provide some
## arguments of its own if it wants.)  It should set `mode' to indicate how
## the chain should be finished.
##
## goto		If no networks matched, then issue a final `goto' to the
##		chain named by the variable `fail'.
##
## call		Run `$finish CHAIN' to write final rules to the named CHAIN
##		(which may be suffixed from the original BASE argument if
##		this was necessary).  This function will arrange to call
##		these rules if no networks match.
##
## ret		If no network matches then return (maybe by falling off the
##		end of the chain).
matchnets () {
  local opt win flags prepare base suffix next net lose splitp
  opt=$1 win=$2 flags=$3 prepare=$4 base=$5 suffix=$6 next=$7 net=$8
  shift 8

  ## If this is the default network, then set the flag.
  case "$net" in default) flags=${flags}default: ;; esac

  ## Do an initial pass over the addresses to see whether there are any
  ## negative ranges.  If so, we'll need to split.  See also the standard
  ## joke about soup.
  splitp=nil
  eval "addrs=\"\$net_inet_$net \$net_inet6_$net\""
  for a in $addrs; do case $a in !*) splitp=t; break ;; esac; done

  trace "MATCHNETS [splitp $splitp] $opt $win $flags [$prepare] $base $suffix $next : $net $*"

  ## Work out how to handle matches against negative address ranges.  If this
  ## is the last network, invoke the PREPARE hook to find out.  Otherwise, if
  ## we have to split the chain, recursively build the target here.
  case $splitp,$# in
    t,0 | nil,0)
      $prepare $flags
      case $splitp,$mode in
	*,goto)
	  lose="-g $fail"
	  ;;
	*,ret)
	  lose="-j RETURN"
	  ;;
	t,call)
	  clearchain mangle:$base#$next
	  lose="-g $base#$next"
	  ;;
	nil,call)
	  ;;
      esac
      ;;
    t,*)
      clearchain mangle:$base#$next
      matchnets $opt $win $flags "$prepare" \
	$base \#$next $(( $next + 1 )) "$@"
      lose="-g $base#$next" mode=goto
      ;;
    *)
      mode=continue
      ;;
  esac

  ## Populate the chain with rules to match the necessary networks.
  eval addr=\$net_inet_$net addr6=\$net_inet6_$net class=\$net_class_$net
  for a in $addr; do
    case $a in
      !*) run iptables -t mangle -A $base$suffix $lose $opt ${a#!} ;;
      *) run iptables -t mangle -A $base$suffix -g $win-$class $opt $a ;;
    esac
  done
  for a in $addr6; do
    case $a in
      !*) run ip6tables -t mangle -A $base$suffix $lose $opt ${a#!} ;;
      *) run ip6tables -t mangle -A $base$suffix -g $win-$class $opt $a ;;
    esac
  done

  ## Wrap up the chain appropriately.  If we didn't split and there are more
  ## networks to handle then append the necessary rules now.  (If we did
  ## split, then we already wrote the rules for them above.)  If there are no
  ## more networks then consult the `mode' setting to find out what to do.
  case $splitp,$#,$mode in
    *,0,ret) ;;
    *,*,goto) run ip46tables -t mangle -A $base$suffix $lose ;;
    t,0,call) $finish $base#$next ;;
    nil,0,call) $finish $base$suffix ;;
    nil,*,*)
      matchnets $opt $win $flags "$prepare" $base "$suffix" $next "$@"
      ;;
  esac
}

## net_interfaces HOST NET
##
## Determine the interfaces on which packets may plausibly arrive from the
## named NET.  Returns `-' if no such interface exists.
##
## This algorithm is not very clever.  It's just about barely good enough to
## deduce transitivity through a simple routed network; with complicated
## networks, it will undoubtedly give wrong answers.  Check the results
## carefully, and, if necessary, list the connectivity explicitly; use the
## special interface `-' for networks you know shouldn't send packets to a
## host.
net_interfaces () {
  host=$1 startnet=$2

  ## Determine the locally attached networks.
  targets=:
  eval ii=\$host_ifaces_$host
  for i in $ii; do targets=$targets$i:; done

  ## Determine the transitivity.
  seen=:
  nets=$startnet
  while :; do

    ## First pass.  Determine whether any of the networks we're considering
    ## are in the target set.  If they are, then return the corresponding
    ## interfaces.
    found=""
    for net in $nets; do
      tg=$targets
      while :; do
	any=nil
	case $tg in
	  *"=$net:"*)
	    n=${tg%=$net:*}; tg=${n%:*}:; n=${n##*:}
	    addword found $n
	    any=t
	    ;;
	esac
	case $any in nil) break ;; esac
      done
    done
    case "$found" in ?*) echo $found; return ;; esac

    ## No joy.  Determine the set of networks which (a) these ones can
    ## forward to, and (b) that we've not considered already.  These are the
    ## nets we'll consider next time around.
    nextnets=""
    any=nil
    for net in $nets; do
      eval via=\$net_via_$net
      for n in $via; do
	case $seen in *":$n:"*) continue ;; esac
	seen=$seen$n:
	eval noxit=\$net_noxit_$n
	case " $noxit " in *" $startnet "*) continue ;; esac
	case " $nextnets " in
	  *" $n "*) ;;
	  *) addword nextnets $n; any=t ;;
	esac
      done
    done

    ## If we've run out of networks then there's no reachability.  Return a
    ## failure.
    case $any in nil) echo -; return ;; esac
    nets=$nextnets
  done
}

m4_divert(-1)
###----- That's all, folks --------------------------------------------------