[Cerowrt-devel] DNSSEC & NTP Bootstrapping

[David Personette]

Phil,

With the exception of the extra dependencies (dig and python), I like this.
I would suggest that if DNSSEC will be enabled, that nslookup (I think
that's the only command line resolver included by CeroWRT/OpenWRT base
installs) be extended to have a similar option as dig, to resolve without
DNSSEC.

The only other issue I see is if the router is brought online before
internet access is available. If I read your code correctly, it will try 4
times per defined server (with and without DNSSEC for IPv4 and IPv6), then
exit. It either needs to keep trying until it succeeds, or be called every
time a connection comes up (shutting down NTPd prior and restarting after).

Thanks.

-- 
David P.



On Mon, Mar 24, 2014 at 3:12 PM, Phil Pennock <
cerowrt-devel+phil at spodhuis.org> wrote:

> On 2014-03-21 at 23:33 -0400, Joseph Swick wrote:
> > I've been lurking for several months now on the list and I remember some
> > discussion about trying to find acceptable methods for bootstrapping the
> > local system time so that DNSSEC would work.
>
> I raised this on the ntp-pool mailing-lists last year, looking for a
> solution because of the chicken/egg bootstrap, with suggested approaches
> and some trial scripts.  Eg:
>
>   http://lists.ntp.org/pipermail/pool/2013-July/006569.html
>
> For context, I'm currently running OpenWRT; attached is the
> /etc/init.d/ntpdate which I'm using.  It relies upon having Python and
> dig installed, as I haven't gotten around to building a small C utility
> to do just this task, but perhaps the approach is useful enough that
> someone else might do so?
>
> In summary: if the current time is less than the timestamp on the
> unbound-maintained copy of the root zone trust anchors, then bump the
> time up at least that far, because we must be at >= that timestamp, and
> this increases the odds that DNSSEC will validate if we haven't been
> off-line for too long.
>
> Then, for each hostname in the $STEP_SERVERS list (which could be
> taken from ntp.conf or uci config or whatever, but here is just
> hardcoded), I try to resolve IPv4 then IPv6, first with DNSSEC left
> enabled, and then with DNSSEC disabled via `dig +cd`.  The first dig
> command to return results is the one which is used.
>
> The idea is to minimize the potential vulnerability of syncing to a bad
> timesource, by using DNSSEC if it's available and works, after making
> sure it has a reasonable chance of working if we've just rebooted, and
> only if we've been off-line for some time do we fall back to insecure
> DNS.
>
> Make sure that the START value is appropriate for your systems; I've
> found the OpenWRT defaults to be sufficiently broken that I stomp on
> them on reinstall.  I run ntpdate once the network and firewall are up,
> but just before ntpd and both of those well before other network
> services which might depend upon time.
>
> Regards,
> -Phil
>
> #!/bin/sh /etc/rc.common
> # Copyright (C) 2006-2008 OpenWrt.org
> # Copyright (C) 2013 Phil Pennock
>
> START=60
>
> STEP_SERVERS="0.openwrt.pool.ntp.org 1.openwrt.pool.ntp.org
> 2.openwrt.pool.ntp.org"
> TIMEOUT="2" # in seconds
> PRESEED_TIMESTAMP_FN="/etc/unbound/runtime/root.autokey"
>
> # The core problem is that with DNSSEC, an invalid time prevents resolution
> # of DNS, but we need DNS to be able to find time-servers to get a good
> time
> # to be able to resolve DNS.
> #
> # We break out of this "Catch 22" situation by _trying_ normal DNS
> resolution,
> # IPv4 and then IPv6, and only if those fail do we forcibly disable DNSSEC
> # by using dig(1)'s +cd flag ("checking disabled"); trying normally first
> # protects us against malicious DNS trying to point us to bad time-servers,
> # if we've enough state that we _should_ already be protected.
> #
> # The "insecure" approach we regress to, as a last resort, is the same way
> # the Internet functioned for decades.  There is a DoS+hijack attack path
> # here, but if we don't have a good battery-backed clock to protect us, we
> # don't have a better solution.
>
> # Also, per a suggestion from Doug Calvert, we can use the timestamp of
> # modification of the unbound root.key file itself as an approximate time.
> # Unbound updates the file on every refresh, so it's not too far off.
>
> preseed_approximate_time() {
>         # Unfortunately, date(1) on OpenWRT can't parse the timestamp
>         # output from ls.
>         python -c '
> import os, time, sys
> fn=sys.argv[1]
> min_time=os.stat(fn).st_ctime
> if time.time() < min_time:
>   want=time.strftime("%Y%m%d%H%M.%S", time.gmtime(min_time))
>   os.system("date -u -s %s" % want)' "$PRESEED_TIMESTAMP_FN" > /dev/null
> }
>
> resolve_hostname_v4() {
> # we use the grep both to filter out cname referrals and to detect empty
> # results
>         local hn="$1"
>         shift
>         dig +nodnssec +short "$@" -t a "$hn" | grep '^[0-9][0-9.]*$'
> }
>
> resolve_hostname_v6() {
>         local hn="$1"
>         shift
>         dig +nodnssec +short "$@" -t aaaa "$hn" | grep -i
> '^[0-9a-f][0-9a-f.:]*$'
> }
>
> resolve_one_server() {
>         local hn="$1"
>         resolve_hostname_v4 $hn && return
>         resolve_hostname_v6 $hn && return
>         resolve_hostname_v4 $hn +cd && return
>         resolve_hostname_v6 $hn +cd && return
> }
>
> resolve_step_servers() {
>         local server ips
>         for server in $STEP_SERVERS ; do
>                 resolve_one_server $server
>         done
> }
>
> start() {
>         preseed_approximate_time
>         for s in $(resolve_step_servers) ; do
>                 /usr/sbin/ntpdate -s -b -u -t "$TIMEOUT" "$s" && break
>         done
> }
>
> _______________________________________________
> Cerowrt-devel mailing list
> Cerowrt-devel at lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/cerowrt-devel
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.bufferbloat.net/pipermail/cerowrt-devel/attachments/20140324/539bc4c0/attachment-0002.html>