I looked at the same thing :) The RTP header is pretty short. You've
technically got 12 bytes of RTP header, but some of it is sequence
number and timestamp which I don't know how you could match on. You can
make reasonable guesses about what the first 16 bits are going to be,
after that you really only know the length of the rest of the header and
the length of the payload. I personally could not write a regexp that
would be substantially more accurate than matching the whole packet
length. Maybe ask whoever wrote wireshark.
I saw this L7 match rule somewhere:
^\x80[\x01-"`-\x7f\x80-\xa2\xe0-\xff]?..........*\x80
but the plain english version of that is something like, "the sequence
starts with 0x80, then there's another byte that might match several
patterns, and then there are some more bytes, and then another 0x80".
Which I don't think is much better than just matching the length. It
also doesn't match any of my actual VoIP traffic anyway.
Here's another thought: You can easily match SIP packets. They contain
plain text headers that you can match. You could just match port 5060,
but Google Voice and 8x8 both stopped using the standard port, so
instead you look for content containing SIP Invite or some such. You
could add matching IP addresses to an address list with a 1 hour
expiration, then subsequent rules look for the appropriately sized UDP
packets to and from the IP addresses that you previously saw sending SIP
packets.
/ip firewall mangle
add action=add-src-to-address-list address-list=recentSIPInvite \
address-list-timeout=1h chain=prerouting content=INVITE disabled=no \
protocol=udp
add action=accept chain=prerouting disabled=no packet-size=200
protocol=udp \
src-address-list=recentSIPInvite
add action=accept chain=prerouting disabled=no dst-address-list=\
recentSIPInvite packet-size=200 protocol=udp
The above matches all the RTP traffic in my office. I could probably be
more accurate than watching for the string "INVITE" and I could narrow
the field of packets that are worth spending CPU on content
matching.....but it works.
Is there no way to make an L7 rule to do this?
Here's what Wireshark gives me. Can't I do something with this
information?
On 9/30/2014 4:05 PM, Adam Moffett via Af wrote:
So one simple workaround I saw was was to match UDP packets at
specific sizes. Every single G.711 RTP packet my phone sends is a
UDP packet that's 200 bytes every time. I'm sure it would eventually
overmatch something, but it's simple and low cost. The example I saw
was matching every UDP packet from 100-400 bytes that wasn't already
matched by some other criteria. A quick test here worked with a rule
set to match at exactly 200 bytes....so I'm thinking rather than
saying every mid size UDP packet is VoIP that maybe I'll match the
specific sizes of packets in common codecs with 20ms frame sizes.
Yeah, it's more complicated than just giving the PBX a public and
setting DSCP on every packet destined to that IP. I can't have
everything end up in the HP queue, that will create more problems
than the one I'm trying to solve. I need to figure out a way to
identify the traffic and mangle DSCP on at the edge routers.
On 9/30/2014 11:04 AM, Adam Moffett via Af wrote:
....but like George, I would also be interested in some sort of
rule that would match RTP voice traffic. I don't see any easy way
to do it, but wireshark seems to pick up on it reliably, so I guess
there's a way.
+1
Seems like the easiest answer.
On 9/30/2014 11:05 AM, Ken Hohhof via Af wrote:
Is the company’s PBX behind their firewall? If you give it a
dedicated IP address, then you can tag based on destination IP.
*From:* George Skorup (Cyber Broadcasting) via Af
<mailto:[email protected]>
*Sent:* Tuesday, September 30, 2014 9:55 AM
*To:* [email protected] <mailto:[email protected]>
*Subject:* Re: [AFMUG] DiffServ and the internet
These are business customers with on-site PBXs with a VoIP
Innovations SIP trunk. Yeah, if we were running a local switch,
then this problem would be a whole lot easier to solve, but
that's not what I have to work with at this point.
As far as I can tell, there's no easy way to identify the VoIP
Innovations audio streams. They come from tons of different
source address and use random ports from 10000 to 20000. And it's
a mix of G711 and G729.
No idea, I'm just the network guy.
On 9/30/2014 9:28 AM, Ken Hohhof via Af wrote:
That’s what I do, there’s another way? We put customer ATAs on
private IPs so it wouldn’t work if traffic bypassed our server.
Is there a configuration parameter on the SIP trunk that tells
it to send RTP traffic directly to the endpoint?
We also have a multisite business customer that uses a hosted
VoIP service (Star2Star) with an appliance at each site, we give
each appliance its own public IP and tag traffic to those IPs.
*From:* Adam Moffett via Af <mailto:[email protected]>
*Sent:* Tuesday, September 30, 2014 9:03 AM
*To:* [email protected] <mailto:[email protected]>
*Subject:* Re: [AFMUG] DiffServ and the internet
I've been cheating up until this point. If you force the audio
to be bridged through your own server then you can tag all the
traffic that goes to and from that server. It doesn't seem to
make a huge difference versus having RTP go straight to the
carrier. If you're not transcoding then the added CPU usage is
minimal. Faxing seems to work better if I'm not bridging the
audio, but why am I faxing anyway, right?
I tried all kinds of stuff tonight, none were any good. I
wonder if there's a way on MT to snoop SIP messages and look
for the SIP contact IPs and mark those. Seems tricky. And I R
no smrt enuf.
On 9/29/2014 9:37 PM, Chris Fabien via Af wrote:
Packet size and rate is pretty consistent right? Just a
thought...
On Mon, Sep 29, 2014 at 8:05 PM, George Skorup (Cyber
Broadcasting) via Af <[email protected] <mailto:[email protected]>> wrote:
Speaking of DSCP and carriers zeroing it in the middle, I
have some VoIP Innovations trunks. I know where the SIP
messages are coming from, so I can mangle a DSCP value
back onto those packets at ingress. But the RTP traffic
comes from all over the freakin place, tons of different
source address, never the same. I've asked if they could
provide a list and pretty much got a no.
Anybody have any ideas? Any way for a MT to identify an
RTP stream and then dynamically add a mangle rule to
change the DSCP value? My MT script-fu is not strong.
