Recently I have seen many attempts to break in to VoIP switches I work with.
So I thought this might be a good time to discuss everyone's security best
practices. If you could relate them to how you implemented them in
FreeSWITCH, it would give us beginners good ideas on how to follow the
principle of "security first". Let me start off the discussion with some
practices we use and have seen in carriers we deal with.

Since we only do carrier-to-carrier stuff, it is easy to restrict the IPs
where we accept traffic from. This is done using either a dedicated firewall
(we use Cisco and pfSense) or host-based (since we use Linux, iptables it
is). We restrict ingress traffic on port 5060 (SIP) to trusted sources only.
Since a lot of carriers are starting to use re-invites and most do not
provide ranges of media IPs they will use, we leave open all UDP ports used
by our servers for RTP, usually between 10,000 and 50,000. In this
situation, how well does FreeSWITCH cope with attacks? For example, if RTP
ports are 10,000 to 20,000, does FS recognize invalid RTP packets and knows
how to deal with them?

Another scenario is user registration of softphones or ATAs, etc. A practice
we use in our lab environment (we will soon be putting servers in
production) is to use lengthy passwords, at least 15 characters, mixing
alpha-numeric characters. We also do not use 4-digit extensions (default in
FS) but use 6-digit. It wasn't started as a security measure but turned out
that many attackers we saw usually try 4-digit extensions. So we are safer
today (maybe not so in the future when attackers wise up). How do you manage
your users, extensions, registrations, etc., from a security aspect?

In the same context, we recently saw a flood of invalid user registration
requests, at close to 60-80 requests per second, for a sustained period of
less than a minute. One proprietary solution we use was unable to handle
this flood of requests and dropped 70% of its call before the attacks
stopped and it was able to recover. Let's say we use top-of-the-line servers
to run FS, maybe two quad-cores with 16GB memory, then should it increase
the ability of FS to deal with such unwanted floods? I assume FS is
configured with default settings. The real question is: does throwing
hardware at the problem solve the issue for FS? And even if it does, is
there anything we can do at the FS level to allow it to handle even more
flooded traffic on its own, without relying on superior hardware?

Then come Man in the Middle attacks. So far I haven't seen a real push in
the industry to mitigate these. One common practice is to use prefixes to
identify traffic from a source. For example, we can use 999912125550000
where 9999 identifies us to the egress gateway and the rest is the number to
call. Of course, since signaling is in plain text someone can analyze this
data and then start using the same prefix to send traffic to our egress
gateway/carrier. They believe the traffic is from us and charge us for the
calls.

A second option is to use SIPS but I haven't seen it being used at all. Of
course, now we are talking about using TCP instead of UDP, bringing its own
advantages and disadvantages.

A third option I am starting to see if the use of IPSec tunnels for
signaling and RTP is exchanged outside of the tunnel. This works well for
carrier-to-carrier traffic but cannot be deployed widely cost-effectively if
you do home user to service provider traffic.

If you have dealt with any one or all of these scenarios, or have come
across others, please share your experiences and solutions with all of us.
Of course, if you can also advise how to implement your security best
practices in FS it would be just wonderful.

What carriers are you using that utilize re-INVITEs?

--
Kristian Kielhofner http://blog.krisk.org

------------------------------
*From*: freeswitch-users-bounces*******<
freeswitch-users-bounces*******>
*To*: freeswitch-users*******<
freeswitch-users*******>
*Sent*: Tue Jun 15 14:37:02 2010
*Subject*: [Freeswitch-users] Your Security Best Practices

Recently I have seen many attempts to break in to VoIP switches I work with.
So I thought this might be a good time to discuss everyone's security best
practices. If you could relate them to how you implemented them in
FreeSWITCH, it would give us beginners good ideas on how to follow the
principle of "security first". Let me start off the discussion with some
practices we use and have seen in carriers we deal with.

Since we only do carrier-to-carrier stuff, it is easy to restrict the IPs
where we accept traffic from. This is done using either a dedicated firewall
(we use Cisco and pfSense) or host-based (since we use Linux, iptables it
is). We restrict ingress traffic on port 5060 (SIP) to trusted sources only.
Since a lot of carriers are starting to use re-invites and most do not
provide ranges of media IPs they will use, we leave open all UDP ports used
by our servers for RTP, usually between 10,000 and 50,000. In this
situation, how well does FreeSWITCH cope with attacks? For example, if RTP
ports are 10,000 to 20,000, does FS recognize invalid RTP packets and knows
how to deal with them?

Another scenario is user registration of softphones or ATAs, etc. A practice
we use in our lab environment (we will soon be putting servers in
production) is to use lengthy passwords, at least 15 characters, mixing
alpha-numeric characters. We also do not use 4-digit extensions (default in
FS) but use 6-digit. It wasn't started as a security measure but turned out
that many attackers we saw usually try 4-digit extensions. So we are safer
today (maybe not so in the future when attackers wise up). How do you manage
your users, extensions, registrations, etc., from a security aspect?

In the same context, we recently saw a flood of invalid user registration
requests, at close to 60-80 requests per second, for a sustained period of
less than a minute. One proprietary solution we use was unable to handle
this flood of requests and dropped 70% of its call before the attacks
stopped and it was able to recover. Let's say we use top-of-the-line servers
to run FS, maybe two quad-cores with 16GB memory, then should it increase
the ability of FS to deal with such unwanted floods? I assume FS is
configured with default settings. The real question is: does throwing
hardware at the problem solve the issue for FS? And even if it does, is
there anything we can do at the FS level to allow it to handle even more
flooded traffic on its own, without relying on superior hardware?

Then come Man in the Middle attacks. So far I haven't seen a real push in
the industry to mitigate these. One common practice is to use prefixes to
identify traffic from a source. For example, we can use 999912125550000
where 9999 identifies us to the egress gateway and the rest is the number to
call. Of course, since signaling is in plain text someone can analyze this
data and then start using the same prefix to send traffic to our egress
gateway/carrier. They believe the traffic is from us and charge us for the
calls.

A second option is to use SIPS but I haven't seen it being used at all. Of
course, now we are talking about using TCP instead of UDP, bringing its own
advantages and disadvantages.

A third option I am starting to see if the use of IPSec tunnels for
signaling and RTP is exchanged outside of the tunnel. This works well for
carrier-to-carrier traffic but cannot be deployed widely cost-effectively if
you do home user to service provider traffic.

If you have dealt with any one or all of these scenarios, or have come
across others, please share your experiences and solutions with all of us.
Of course, if you can also advise how to implement your security best
practices in FS it would be just wonderful.

I would also be interested to see some more best practices

Recently I have seen many attempts to break in to VoIP switches I work with.
So I thought this might be a good time to discuss everyone's security best
practices. If you could relate them to how you implemented them in
FreeSWITCH, it would give us beginners good ideas on how to follow the
principle of "security first". Let me start off the discussion with some
practices we use and have seen in carriers we deal with.

Since we only do carrier-to-carrier stuff, it is easy to restrict the IPs
where we accept traffic from. This is done using either a dedicated firewall
(we use Cisco and pfSense) or host-based (since we use Linux, iptables it
is). We restrict ingress traffic on port 5060 (SIP) to trusted sources only.
Since a lot of carriers are starting to use re-invites and most do not
provide ranges of media IPs they will use, we leave open all UDP ports used
by our servers for RTP, usually between 10,000 and 50,000. In this
situation, how well does FreeSWITCH cope with attacks? For example, if RTP
ports are 10,000 to 20,000, does FS recognize invalid RTP packets and knows
how to deal with them?

Another scenario is user registration of softphones or ATAs, etc. A practice
we use in our lab environment (we will soon be putting servers in
production) is to use lengthy passwords, at least 15 characters, mixing
alpha-numeric characters. We also do not use 4-digit extensions (default in
FS) but use 6-digit. It wasn't started as a security measure but turned out
that many attackers we saw usually try 4-digit extensions. So we are safer
today (maybe not so in the future when attackers wise up). How do you manage
your users, extensions, registrations, etc., from a security aspect?

In the same context, we recently saw a flood of invalid user registration
requests, at close to 60-80 requests per second, for a sustained period of
less than a minute. One proprietary solution we use was unable to handle
this flood of requests and dropped 70% of its call before the attacks
stopped and it was able to recover. Let's say we use top-of-the-line servers
to run FS, maybe two quad-cores with 16GB memory, then should it increase
the ability of FS to deal with such unwanted floods? I assume FS is
configured with default settings. The real question is: does throwing
hardware at the problem solve the issue for FS? And even if it does, is
there anything we can do at the FS level to allow it to handle even more
flooded traffic on its own, without relying on superior hardware?

Then come Man in the Middle attacks. So far I haven't seen a real push in
the industry to mitigate these. One common practice is to use prefixes to
identify traffic from a source. For example, we can use 999912125550000
where 9999 identifies us to the egress gateway and the rest is the number to
call. Of course, since signaling is in plain text someone can analyze this
data and then start using the same prefix to send traffic to our egress
gateway/carrier. They believe the traffic is from us and charge us for the
calls.

A second option is to use SIPS but I haven't seen it being used at all. Of
course, now we are talking about using TCP instead of UDP, bringing its own
advantages and disadvantages.

A third option I am starting to see if the use of IPSec tunnels for
signaling and RTP is exchanged outside of the tunnel. This works well for
carrier-to-carrier traffic but cannot be deployed widely cost-effectively if
you do home user to service provider traffic.

If you have dealt with any one or all of these scenarios, or have come
across others, please share your experiences and solutions with all of us.
Of course, if you can also advise how to implement your security best
practices in FS it would be just wonderful.

Let me put it this way: of more than 40 carriers I work with, only six have
some sort of problem if they see re-invite. Some of the bigger names most
people recognize all support re-invite (and some even prefer it): AT&T,
Qwest, and Verizon, among others.

Let me clarify when I say re-invite. We are basically talking about keeping
yourself in the signaling path but once the call is answered to invite your
ingress and egress carriers to use each others' media IP instead of your
media IP. This way you don't have to deal with media once call is connected.

We have bypass media.

/b

I believe it is nice to say as well that most small transit carriers see the
bypass media as a commercial threat since they will disclose in part of the
packet to which IP address they are actually terminating the call to.

I've been through this with lots of carriers that do nothing but LCR dips
and transit.

JM

This is what is wrong with this industry.  You have every Tom, Dick and Hary setting up Asterisk and FreeSWITCH boxes without an ounce of clue what he heck to do. Then you have mom and pop shops that provide really sub parservices by routing media all over the globe when not needed.  This result in software such as Asterisk or FreeSWITCH getting a bad name when the meia goes to hell because TomDickHarry's voip service he runs on his cable mdem went out because it rained outside.

So to summarize, either do it right or go home.

/b

We have some parameters to disable some commonly abused features in a
carrier env.

in the sofia profile

set manage-presence to false

If you are not providing registration or client transfers disable them
completely
set disable-transfer to true
set disable-register to true

And register for ClueCon where someone will probably give a talk on VoIP
security!

--
Anthony Minessale II

FreeSWITCH  http://www.freeswitch.org/
ClueCon  http://www.cluecon.com/
Twitter:  http://twitter.com/FreeSWITCH_wire

AIM: anthm
MSN:anthony_minessale*******>
GTALK/JABBER/PAYPAL:anthony.minessale*******>
IRC: irc.freenode.net #freeswitch

FreeSWITCH Developer Conference
sip:888*******>
googletalk:conf+888*******>
pstn:+19193869900

I was reading up on some security advice and came across an article from
Digium ( http://blogs.digium.com/2009/03/28/sip-security/ ). A few points that
stood out for me were:

"Make your SIP usernames different than your extensions". This sounds like
good advice because now an attacker has to guess the user name and password
instead of just a password. The biggest benefit to this is that even if
someone knows the format of your extension numbers, they are not able to use
it for registration credentials. Of course, the issue of DoS using a large
number of simultaneous authentication requests still remains. How can we
deal with this?

"... reject bad authentication requests on valid usernames with the same
rejection information as with invalid usernames ..." How can we do this in
FS? I haven't looked at this yet so maybe FS does such a thing anyways;
advice from your experience and expertise required.

"Allow only one or two calls at a time per SIP entity, where possible." This
is standard practice in prepaid phone card business. They allow only one
simultaneous call and if you try to use the PIN to make another call it's
not allowed. At the application level we may be able to do this in FS but it
reminds me of another scenario. Let's say a user's registration is
compromised. Can we prevent registration with same credentials from two
different locations in FS? For example, one registration attempt is from New
York and the other from Baltimore. This would affect the legitimate user
because they might not be able to register. Does anyone restrict this way
using FS?

Apart from the article mentioned, there are some other issues, one of which
is ANI authentication. A lot of pinless prepaid phone services use your
caller ID to authenticate and authorize a call. you give them your phone
number and when you call from there you can use the service. With VoIP it's
very easy to spoof ANI. I believe the concept of "context" helps a lot here.
Say you know the ANI's you assigned to your registered users. They would be
in the "default" context, for example. Now if an attacker uses the same ANI
but tries not from a registered endpoint but say from one of your carriers.
His context would be "public", for example. This way the combination of ANI
with context can prevent such attacks.

But what if the authorized ANI (say 4145550000) is not from a registered
endpoint but from a cell phone, similar to the pinless services offered
today. In this scenario I think there's not much we can do. If we have given
a user an access number (say 2125550000) provided to us by a DID provider,
say ABC, then whenever someone dials 2125550000 then we get the call only
through ABC, whom we have allowed. But we can't be sure that 4145550000 is
actually 4145550000 and not someone pretending to be them. Has anyone
figured out a way to detect and/or prevent such fraudulent calls? Is it even
possible?

You really should try FreeSWICH and understand the security.  In the end its YOUR job to secure it... not ours.  I can't stop someone from setting up FreeSWITCH in an insecure way.  Just like you can't stop someone from setting up a CGI script that owns your box.

How about you set the AUTH username different then the username and then have a password too?

All bad requests get a 403.  All requests good or bad get a 401 then a 403 if invalid.  No indications at all that the username is valid.

Mod_limit.

Never a good idea to use caller id as auth.

NO.

/b

Use fail2ban or some other mechanism. If someone is sending mass numbers of
SUBSCRIBEs then you should be handling that the same way you handle all
brute-force attacks - at the firewall. Although you should definitely not
make hacking your VoIP system easy, you should also keep in mind that
external security is an absolute must. Use the layered approach

This I don't know. Is it handled down in Sofia? Brian or Tony could tell us.

I'm sure you could use a combination of ACLs and mod_limit.

I cringe at the thought of doing *ANY* kind of authentication based on
incoming caller ID...

I think this is where mod_cleo comes in. :P Seriously, you probably should
treat inbound calls as untrusted regardless of where they're from, at least
until you've done some level of verification. A PIN code isn't hyper-secure,
but it is better than nothing.

You can't prevent all of these attacks. The best you can hope to do is to
make the attacks as difficult as reasonably possible without irritating your
paying customers.

I invite others to chime in. Also, I've started an informal security best
practices page on the wiki:

http://wiki.freeswitch.org/wiki/Security_bp

Everyone put your thoughts up and we'll formalize it at a later time.
-MC

FYI, there are soft phones out there that FAIL at this:
wxCommunicator
GoldMine CRM Softphone (Front Range Solutions)

These two phones do not support the concept of "auth username" - they ASSUME
that the username is the auth username. Naughty naughty.
-MC

The security related problems were discussed on almost every soft switch
mailing lists I'm watching. And the outcome was (in plain English which
Americans do not understand) - never provide working configuration samples in
the softswitch distribution :-( Let the end user build the configuration.

I don't know that I agree with such an extreme view. Better to give them an
example of a locked-down config than to let them flounder on their own and
enterprise *SHOULD* consult a security professional to make sure that
everything on their LAN/WAN is up to snuff.

-MC

Give them a non-working example. And provide a comment on almost every line
why this will not work. Let the customers exercise their brain first. You don't
need dumb customers, am I right? You'll see a good outcome.

Sorry, but I can't resist. It's my daily work to get rid of unwanted
customers. "Unwanted Customers" doesn't sound good in any language. But it's a
reality...