Outbound filtering of Web requests using Squid as a Proxy server

Frequently in my line of work I’ll be asked about filtering of outbound traffic from application servers. There are two schools of thought here, one is that an app server can have unfiltered access to the internet, and the other that the app server should have as little access to any resources (both inside and outside of the solution) as needed to preform its role.

This generally isn’t an issue if site to site VPNs, static IPs or similar are being used on the destination side. But what happens if your application requires access to something like Youtube, Facebook or Flickr. As these cloud services are not managed by the customer, we have no idea if they are on static IP addresses (and in the case of flickr, they do seem to change moderately frequently).

With this in mind a traditional Layer3/Layer4 firewall is only going to be able to handle this if it supports DNS resolution in its access-list set, and unfortunately (but for good reason) this is not a common feature. Cisco did introduce this to the ASA firewalls in 8.4, however I personally have not used this, so at the moment its still a bit of an unknown and I can’t recommend it to a customer.

There is however another way of doing this, whilst it might not be a perfect situation, it does at least allow you to filter outbound traffic.

The Squid proxy server has been around for quite some time and is quite a stable product, both in the forward (outbound) and reverse (inbound) HTTP proxy space. We’re going to use this to preform our outbound proxying. It is possible to use commercial products like a BlueCoat proxy, however I’m going to concentrate on the FOSS solution here.


Before we start we need to have the following:

  • A Linux Server (for this example I’m going to be using CentOS 6.4, however any linux distribution should work)

Installing Squid

This is a really simple task on most linux distributions, as not only has squid been since the early 90’s, it’s also really popular! You can use the package manager to install squid on most distributions

You should get a response similar to below:

We now would need to configure squid to start on boot

 SSL Proxying

Squid has a rather nice feature called SSLBump which allows us to preform a Man In the Middle SSL Proxy. Privacy issues aside on this feature (after all we’re using it for servers not for end users) this is going to work for us from the server side of things. One key thing to note is we have to trust the CA, that we’re going to generate, on all applications / servers. I’m not going to cover how to do this in this post.

Normally when we create an SSL certificate we’d do this with a specific domain, however as we’re going to be proxying for all domains we’re going to use a wildcard certificate. For the “Common Name” or Server name, we need to chose “*” as the value.

In order to create the CA you can follow the following post. One point of note is to ensure that you do not do this on the Squid server, as this would mean that should the server be compromised, the CA (which is trusted on multiple servers) is now trusted as well.

We need to create the certificate using the CA script as per the above post. CA -newreq This will look similar to

Once this is completed you’ll need to sign this with the  CA -sign command

Once this is completed, ensure that newcert.pem and newkey.pem are copied to the squid server. You will then also need to remove the passphrase from the key.

Once this is done, you’ll need to then also copy the cert into the same file

Configuring Squid

We’re going to make a very simple squid config, allowing access from the App servers to youtube.com, but no other hosts. Replace  /etc/squid/squid.conf with the following


Testing Squid

We’re going to use the curl command to test that the ACLs are working

First lets test google, this should fail. We specify the proxy with the -x flag

As you can see we get a 403 on this from Squid

Lets now try http access to youtube.com

This works as expected. Lets try https to youtube.com now!

This has failed as we’ve not got the CA certificate in the bundle that curl uses, lets get curl to ignore the SSL certificate

Now lets just make sure that other https sites don’t work.

 Forwarding all traffic via the Proxy server

Now the way that this is done depends on the firewall or router in use. What we need to achieve is to either D-NAT or redirect all traffic on port 80 / 443 outbound to the Squid server.

For a Cisco ASA there is a guide on how to do this with WCCP

For a Linux based device you would want to have a IPTables rule similar to


Creating a CA using OpenSSL – with OCSP

SSL Certificates are a source of huge amounts of confusion. There are two things that a SSL session will provide. The first is encryption, which can be provided with “self signed” certificates. The second, and arguably the more important is authentication of the remote server. This is managed by “Certification Authorities”. Web Browsers will have a set of known CAs that are trusted, and any certificate signed by them is therefore also trusted. Obviously if a CA has had a security breach then all bets are off.

Within an organisation it is usually preferable for NON PUBLIC facing sites and services to use ‘self signed’ or internal CA signed certificates. The later is usually more sensible, however it comes with the issue of more administrative time is required, and also that all clients must trust this CA.

There are various different ways of creating a CA, Windows Server 2003 and above come with their own CA software, and most UNIX/Linux distributions have OpenSSL available.

In this guide I’m going to walk through the creation of a CA using OpenSSL. I’m also going to look at enabling additional features such as OSCP (a way of clients confirming if a certificate is still valid) and go over how to create “Server Alternative Name” certificates (also known as UC or SAN certs, allowing multiple hostnames/domainnames to exist on the same cert).

One key thing to remember here is security of the CA. You must ensure that no unauthorized access is permitted to the CA, as if someone has been able to gain this, they will have access to issue certificates.

I’m also going to ensure that we setup OCSP, which is a way of clients checking to see that certificates are still valid and not revoked.


Before we start we need to have the following:

  • A Linux Server with openssl installed (for this example I’m going to be using CentOS 6.4, however any linux distribution should work)
  • A Domain name (in this example I’m going to use test.local)

Configure OpenSSL

On a CentOS/RedHat system there is already a basic openssl.cnf file created, that the scripts for managing a CA already take into account.  /etc/pki/tls/openssl.cnf

Open this up in which ever editor you like and do the following:

  • Locate  countryName_default = XX and change the XX to which ever country code you are in, for example the United Kingdom would be GB
  • Locate  #stateOrProvinceName_default = Default Province and edit this line so there is no # at the start, and that Default Province now is set to our State/Proviince/County/City
  • Locate  localityName_default = Default City and edit this to be your City
  • Locate 0.organizationName_default  and edit this to be your City

At this point we’ve edited the config so that for any new requests you won’t have to type these in!

Whilst still in the Text Editor we need to setup the OCSP side of things.

  • Locate the  [ usr_cert ] section and add 

    In this example I’m going to put this on the CA, but this is *NOT* a good idea from a security perspective. You want the CA to have as little (if indeed any) access from the outside.
  • We also need to create the OCSP ‘extensions’ section. Add this to the end of the file


Create the CA

We’re going to use the OpenSSL CA script to do this.

  • Change directory to  /etc/pki/tls/misc 
  • Run the CA command:  ./CA -newca 
  • Whilst Running it you will be asked
              • File name : Just hit enter here
              • PEM Passphrase  : this is the password you will use for the CA. Make sure it’s secure!
              • Country Name : Hit enter here
              • State or Province Name : Hit enter here
              • Locality Name (eg, city) : Hit enter here
              • Organization Name (eg, company) : Hit enter here
              • Organizational Unit Name : Hit enter here
              • Common Name (eg, your name or your server's hostname) : For this its generally considered best to set this to ca.domain, so in this case ca.test.local
              • Email Address : Hit enter here
              • A challenge password : Hit Enter here
              • An optional company name : Hit enter here
              • Enter pass phrase for /etc/pki/CA/private/./cakey.pem :: Enter the CA password here

The end output should look similar to


At this point you have a CA setup and ready to go. You will need to ensure that the CA public certificate is installed on the browsers / devices that you will be using. This can be downloaded using a SCP client from  /etc/pki/CA/cacert.pem

Creating a OCSP signing certificate

In order to host an OCSP server, we have to generate a OCSP signing certificate. If you’re going to have multiple OCSP servers, you may want to have multiple certificates.

We’re going to create a directory, and a request for the certificate

At this point we now need to sign the request and make the certificate

You will be asked for

  • CA Passphrase
  • Sign the certificate? [y/n]: Say yes to this
  • 1 out of 1 certificate requests certified, commit? Say yes to this as well

Start OCSP server

At this point we now also need to run the OCSP server. Be aware that this is going to run as root in this example, which you should *NOT* do. You will want to ensure permissions are done in a way that a normal user, I’m not going to cover this at the moment though. Start the server with the following


 Issuing a Certificate

Now that you’ve got a working CA, you can sign any certificate requests. There are multiple ways of creating these, some software will provide you a CSR, but in this example I’m going to do this all on the CA its self (don’t do this in production!)

  • Change directory to /etc/pki/tls/misc
  • Run the CA command: ./CA -newreq

This will give a result similar to below

We now need to sign the certificate

  • Run the CA command 

This will give a result similar to

The certificate now exists and can be seen in newcert.pem (and the key in newkey.pem)

Checking OCSP status

We can now check to see if the above certificate is valid via OCSP:

openssl ocsp -CAfile /etc/pki/CA/cacert.pem -issuer /etc/pki/CA/cacert.pem -cert newcert.pem -url -resp_text

This will return an address similar to below:

 Revoking a certificate.

Oh no! The certificate above has been compromised. We need to revoke it. This isn’t as difficult as you may think. We have a copy of all of the certificates on the CA. If we look at the certificate serial number (c5:07:3c:dc:c5:8a:cb:ad in this case) this file should exist in  /etc/pki/CA/newcerts/ To revoke you need to

  • Revoke the certificate  openssl ca -revoke /etc/pki/CA/newcerts/C5073CDCC58ACBAD.pem 
  • Verify that the certifcate is revoked  openssl ocsp -CAfile /etc/pki/CA/cacert.pem -issuer /etc/pki/CA/cacert.pem -cert /etc/pki/CA/newcerts/C5073CDCC58ACBAD.pem -url -resp_text


RouterBoard as a Home Router – 4 1/2 years on

A while back I mentioned a follow up to an old blog post about the RouterBoard that i’d recently purchased and setup for home use. This is a very belated update on that board.

My requirements have since changed from the original post, but not dramatically so. The requirement for LACP has disapeared, IPSec is no longer used, but a requirement for Dynamic Routing has appeared.

All in all, I have to say that I still cannot recommend RouterOS enough. I’ve been using it the past 4 1/2 years, and have recommended a large number of folks to use it.

The main reason behind this is that it just works, there’s not really any faffing about that needs to be done, and if you’re running the stable release, everything does just work.

Feature wise, this is right up there with some of the big brands (Cisco, Juniper et all), however its fair to say not with the same price tag.

Continue reading RouterBoard as a Home Router – 4 1/2 years on

iSCSI Target re scanning on VMWare

If you’re using iSCSI on VMWare but have a requirement to rescan the luns after a machine has booted (For example a VM which has DirectPath to a Storage card enabled, which is hosting your iSCSI luns) you can simply do so with the following command

F5 LoadBalancing on a per app mountpoint

With some customers solutions I’ve seen a common requirement to do loadbalancing decision based on the actual application server serving the content, this obviosuly introduces a few issues if you’re using a single base URL for this

If we take the example below

With this in mind, its not possible to use traditional Layer 3 / Layer 4 load balancers, and would require a L7 Load balancer, such as a F5 LTM or Riverbed Stingray (ZTM/ZXTM). I’m going to concentrate on the F5 in this example.

On the F5 you have the abbility to use a iRule to preform load balacning actions. On a Virtal Server that has the “http” profile enabled, you would be able to add a iRule similar to below.

There are multiple events that this will trigger.

  2. This event is triggered whenever a new connection is made to the Load balancer. In our case the code will check to see if the virtual servers name contains “testing”, which if it does sets the serverpool variable to contain “testing”, otherwise it will set it to “liveserver”

  4. This event is triggered on any new HTTP Request. In our case this preforms a ‘switch’ (a multiple if/else statement) on the URL. We do however preform two “transformations” on the URL, the first is we convert it to lower case. the second is that we only take the URL between the first two /’s. So for the URL http://www.example.com/app1/test we would use app1 for teh switch statement.
    Based on the path, we will then set the NEWPOOL variable, and then set the pool to NEWPOOL

  6. This event is triggered when the server send a response to a HTTP Request. We add the “X-AP” header to the response, and set this to the NEWPOOL variable.

Yubikey and server authentication

After starting to use the Yubikey for LastPass and various other online servers I’ve started also using my yubikey for SSH access to my server(s).

I’ve touched on google_authenticator and pam_yubico for authentication in a previous post however I will be going into this in a bit more detail.

Taking a machine at home as an example. My requirements are simple

  • NO SSH Key access to be allowed – as there is no way to require a second factor with an SSH Key (Passphrases can be removed or a new key generated)
  • Access from Local machines to be allowed without Two Factor being enabled
  • Yubikey to be the Primary TFA
  • Fall back to google authenticator should either the Yubico servers be down, an issue with my keys or I just don’t have a USB port available (IE I’m on a phone or whatever)
  • In order to meet these requirements I’m going to need the following

  • yubico-pam Yubikey PAM
  • Google Authenticator PAM
  • pam_access
  • The server is running Archlinux, and luckily all of these are within AUR – and as such I’m not going to cover the install of the modules.

    In order to restrict SSHd access as above I need the following auth lines in /etc/pam.d/sshd

    The next step is ensure that the relevant users and IP are listed in /etc/security/access_yubico.conf

    After this is setup we will also need to setup the yubikey file /etc/yubikey

    I’m not going to cover configuration of google authenticator with the google-authenticator command

    The final changes are to the /etc/ssh/sshd_config ensuring that the following are set

    PAM and Two Factor authentication

    As the need for Two factor authentication is a requirement for PCI-DSS (Payment Card Industry standard) and SSH Key with password is not always deemed to be an acceptable form of Two factor authorisation there is now a surge in different forms of two factor auth, all with their own pros and cons.

    For a small business or ‘Prosumer’ (professional consumers) the market incumbent (RSA) is not a viable option due to the price of the tokens and the software / appliance that is required. There are cheaper (or free!) alternatives for which two that I’ve used at Google Authenticator, and Yubikey.

    Google Authenticator is an OATH-TOTP system that much like RSA generates a one time password once every 30 seconds. It’s avaiable as an App for the Big three mobile platforms (iOS, Android and Blackberry).

    Yubikey is a hardware token that emulates a USB keyboard, that when the button is pressed, generates a one time password. This is supported by services such as lastpass.

    Both solutions have the ability to be used with their own PAM modules. Installation of either is simple, but what happens if you want to use both, but only require one of these.

    Luckily PAM makes it quite easy !

    In the above example the user must enter a password and then provide either their yubikey or their google_authenticator.

    Should the password be incorrect the user will still be prompted for their yubikey or google authenticator, but will then fail. Should they provide a password and then their yubikey, they will not be asked for their google authenticator. Should they provide password and not a yubikey, they will be prompted for their google authenticator!

    A quick (and quite unscientific!) break down of Rackspace CloudFiles UK vs Amazon S3 (Ireland)

    (Disclaimer – I’m a Rackspace Employee, the postings on this site are my own, may be bias, and don’t necessarily represent Rackspace’s positions, strategies or opinions. These tests have been preformed independently from my employer by my self)

    As Rackspace have recently launched a ‘beta’ Cloudfiles service within the UK I thought I would run a few tests to compare it to Amazon’s S3 service running from Eire (or Southern Ireland).

    I took a set of files, totalling 18.7GB, with file sizes ranging from between 1kb and 25MB, text files, and contents being mainly Photos (both JPEG and RAW (cannon and nikon), plain text files, GZiped Tarballs and a few Microsoft Word documents just for good measure.

    The following python scripts were used:

    Cloud Files




    The test was preformed from a Linux host which has a 100MBit connection (Uncapped/unthrottled) in London, however the test was also preformed with almost identical results from a machine in Paris (also 100mbit). Tests were also run from other locations (Dallas Fort Worth – Texas, my home ISP (bethere.co.uk)) however these locations were limited to 25mbit and 24mbit , and both reached their maximum speeds. The tests were as follows:

  • Download files from Rackspace Cloudfiles UK (these had been uploaded previously) – This is downloaded directly via the API, NOT via a CDN
  • Upload the same files to S3 Ireland
  • Upload the same files to a new “container” at Rackspace Cloudfiles UK
  • Download the files from S3 Ireland – This is downloaded directly via the API, NOT via a CDN
  • The average speeds for the tests are as follows:
    Download: 90Mbit/s
    Upload: 85MBit/s
    S3 Ireland
    Download: ~40Mbit/s
    Upload : 13Mbit/s


  • Cloud files seems to be able to max out a 100mbit connection for both File
  • S3 seems to have a cap of 13mbit for inbound file transfers?
  • S3 seems to either be extremely unpredictable on file transfer speeds for downloading files via the API, or there is some form of cap after a certain amount of data transferred, or there was congestion on the AWS network
  • Below is a graph showing the different connection speeds achieved using CF & S3

    As mentioned before this is a very unscientific test (and I’d say that these results have not been replicated from as many locations or as many times as I’d like to, so I would take them with a pinch of salt) , but it does appear that Rackspace cloudfiles UK is noticeably faster than S3 Ireland

    IRSSI Prowl Notifications

    A quick script to send notifications from IRSSI for privmessages and also for highlights, I’ll put more commentary on later, but for now..

    RouterBoard as a Home Router – 7 Months on – Part 1

    At the new year I decided that I was fed up with having my main Unix server acting as a Router (amongst other things) and decided to bite the bullet and get a full blown router. Here in lay a dilema. Being the fact that I’m a geek, I couldn’t settle for a “home” unhackable router. So this instantly ruled out most of the commercial available routers, baring those that run OpenWRT. Now don’t get me wrong, OpenWRT is more than capable, but I just didn’t feel like having to worry about hardware support, fighting with IPTables and getting hardware that probally wouldn’t scale. Now before anyone starts thinking “Scaling, but this is for a home connection!”, this is true. However I do sync my DSL at the full  24244 kbps Downstream, and 2550 kbps upstream (I live under 200m from the exchange according to my line attenuation, also my ISP doesn’t bandwidth cap, and allow for FastPath and similar to be enabled. Go BeThere!) . Also at the time, I was seriously considering investing in a secondary connection for additional bandwidth. This meant that I was left with a few choices

    • Build my Own. Using something like an ALIX/Sokeris and use something like FreeBSD (or something with a webgui for when I feel rather lazy, such as m0n0wall or pfsense. Both I’ve used previously with great success)
    • Cisco. Yes, the 800 pound gorrila of home. A ‘cheap’ 1800 or similar was going to set me back about £400, however this would have provided me most of what I needed.
    • RotuerBoard. These where, to me at least, relativly unknown. I originally looked at them for building my own system with them, and then discovered RouterOS came with the boards. This was an instant sale.

    After my first look at RouterOS I was basically sold. Main reasoning behind this was that it was a comercial Linux distribution, that actually worked well as a router, and shipped with both a CLI (Nortel-esq in this case) and a *shock* gui application. It also met my main criteria.

    • Support for 802.1Q. I have multiple vLANs at home so having support for dot1q was a necessity
    • Support for 802.3ad. As I have a few machines connecting via the router, I needed the throughput, as I don’t have gigabit switching LACP support was a necessity.
    • Support for Wireless. All good routers for the home (even a geeky one) need support for 802.11(a/b/g).
    • Support for SubSSIDs. Relating to the above, I didn’t want to have 7 wireless cards for my various networks
    • Support for WPA2-PSK and WPA2-EAP. I use RADIUS to authenticate all my personal stations to a central authentication system, but I don’t want to have to add guests to this, so PSK should also be supported.
    • Support for OpenVPN. I don’t like having my traffic to / from home going in the clear at all, so I needed to be able to connect via a VPN of some sort, My preference is OpenVPN for c2s vpns (s2s is still IPSEC…. which leads onto the next point)
    • Support for IPSec. I connect to various friends networks, and yet again, don’t want this sort of traffic in the clear, we made the standard IPSec (3des/md5) a while back
    • Support for “Unlimted” Firewall rules. This may sound silly, but anyone who has worked with the lowend Sonicwalls will know what I mean, only being able to put 20 rules is EXTREMELY restrictive especially with multiple vlans! (I’ve got roughly 300 rules)
    • Support for setting DHCP options. I used VMWare ESX at home for my test lab, so I require to be able to setup the DHCP server to be able to send the correct options for PXE (or gPXE) so this was a requirement
    • Quick booting. As silly as this may sound, I don’t want boot times of upwards of 30 seconds for my router.
    • Support for Bridging of interfaces with Firewall rules. This one is rather self explanatory really!
    • Support for UPnP. Lets face it, UPnP is required for any form of Voice/Video chat these days over the main IM networks (YIM/AIM/MSNIM)
    • Support for NetFlow or similar. This one is a nice to have, as I like to use flow-tools to generate a rough guess on what type of traffic is flowing through my network
    • Support for Traffic Shaping. Ah yes, the holy grail of routers. Unfortunately the likes of TC on linux requires a degree in astrophysics to get working how you’d like!
    • Easy configuration.

    After discovering (via the x86 installable and the demo units) that RouterOS would let me do all of the above, I decided to give it a whirl.