[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: New Draft: draft-lewis-infrastructure-security-00.txt



I have added a few comments.
Look for djs> to mark my comments.
I will also snip as much of the text as possible leaving only the
related portions to provide context.



<SNIP>

1.  Introduction


 <SNIP>
   This document presents
   techniques that, together with network edge ingress filtering and RFC
   2267 and RFC 3704, create a layered approach for infrastructure
   protection.
djs> create defense in depth approach for infrastructure protection.

   Denial of Service (DoS) attacks are common and the network
   infrastructure itself is a target.   Attacks targeting the network
   infrastructure can take many forms, ranging from bandwidth saturation
   to crafted packets destined to a router.  These attacks might use
   spoofed source address or they might use the true address of source
   of the traffic.  Regardless of the nature of the attack, the network
   infrastructure must be protected from both accidental and intentional
   attacks.
djs> infrastructure must be protected from accidental floods and
intentional attacks.
djs> Additionally this protection will assist in preventing the network
elements from 
djs> being used as reflectors in attacks against others.

   The techniques outlined in this document and described in section 2
   below, provide a layered approach for infrastructure protection:

djs> provide a defense in depth approach for ...

   Edge policy (filtering and precedence), per device traffic policy
   enforcement for packets destined to a device and finally,
   routing/address advertising best practices to limit core network --
   that is P and PE infrastructure -- exposure.

   This document is aimed at network operators who would like to
   "harden" their infrastructure and make it more resilient to external

djs> This document is targeted at network operators who would like to
djs> limit their exposure to risks within their infrastructure and ...


   attack.  These techniques are designed to be used in addition to
   specific protocol or application security features implemented in
   network devices.

   Infrastructure protection is a complex topic, improving protection is
   always beneficial.

<SNIP>


2.1.  Edge Infrastructure Access Control Lists


   Edge infrastructure access control lists are ingress access control
   lists that filter traffic destined to the network only.  They should
   permit all traffic through the network.  
djs> They should permit all transit traffic.

Explicit filtering of
   traffic destined to network devices creates a first level of
   protection at the network edge: only traffic explicitly permitted
djs> traffic destined towards network devices creates a first layer of
djs> protection at the network edge. Only traffic explicity permitted


   into the network can reach a device beyond the PE router with the
   filter.

   Although very effective, edge infrastructure access control lists are
   not perfect and, like any filter lists, must be maintained and

djs> are seldom implemented completely or perfectly and, ...

   updated.  Furthermore, while widespread deployment on ingress
   interface provides the most protection (which in some cases will not
   be possible), some deployment is better than no deployment.
djs> (complete deployment in some cases will not be possible)
djs> partial deployment reduces the exposure of risks to some degree 
djs> depending on how complete the deployment.

djs> The closer to the ingress point the better, however some edge
devices will not support filtering.
djs> In those cases a filter that is close to the edge can still limit
the exposure.




2.2.  Edge Remarking


   We define Edge Remarking as ensuring that ingress IP precedence or
   DSCP values match expected values within the context of security.
   This provides another layer of defense particularly for traffic
   permitted through any of the Edge Infrastructure Access Control



Gill, Lewis, Quinn, Schoenmaker                   Section 2.2.  [Page 5]

INTERNET-DRAFT                  Expires:                       October
2006


   Lists.   In this RFC we focus only on using Edge Remarking best
   practices to enforce security policies.

djs> Remark packets that will traverse network elements that honor IP
precedence or DSCP valuses. 



2.3.  Device and Element Protection


   Each device infrastructure device should enforce local rules for
   traffic destined to the device itself.  These rules can take the form
   of filters (permit/deny) or rate limiting rules that allow ingress
   traffic at specified rates.  These should complement any existing
   Edge Infrastructure Access Control Lists.

   The deployment of these local device protection rules compliments the
   edge techniques by protecting the device from traffic that: i) was
   permitted but violates device policy, ii) could not be filtered at
   the edge, iii) entered the network on an interface that did not have
   ingress filtering enabled.

djs> iii is a subset/element of ii right?




2.4.  Infrastructure Hiding


   Hiding the infrastructure of the network provides an elegant
   mechanism for protecting the network infrastructure.  If the
   destination of an attack is to an infrastructure address that is
   unreachable, attacks become far more difficult.  
djs> If the destination of an attack is towards an infrastructure
address 
djs> that is not externally targetable or has limited targetablity,
attacks become far more difficult.

Infrastructure
   hiding can be achieved in several ways:

          -  MPLS techniques
          -  IGP configuration
          -  Route advertisement control





3.  Edge Infrastructure Access Control Lists


   Edge Infrastructure Access Control Lists (EIACLs) are a specific
   implementation of the more general Ingress Access List.  As opposed
   to generic ingress filtering which denies data (sometimes referred to
   as user) plane traffic, edge infrastructure access control lists do

djs> (sometimes referred to
djs>  as user plane) traffic, edge infrastructure access control lists
do



Gill, Lewis, Quinn, Schoenmaker                     Section 3.  [Page 6]

INTERNET-DRAFT                  Expires:                       October
2006


   not attempt to deny traffic going through the devices, rather this
   form of access control limits traffic destined to infrastructure
djs>not attempt to deny transit traffic, rather this
djs>form of access control limits traffic destined towards
infrastructure


   equipment while permitting -- if needed, explicitly -- traffic
   through the network.




3.1.  Constructing the Access List


   Edge Infrastructure Access Control Lists permit only required traffic
   to the network infrastructure, 
djs>Edge Infrastructure Access Control Lists permits only required
traffic
djs>targeted towards the network infrastructure,

   while allowing data plane traffic to
   flow through unaffected. 
djs>while allowing data plane traffic to
djs>flow through unfiltered. 


   The basic premise of EIACLs is that only a
   relatively limited subset of traffic, sourced from outside your AS,

djs>AS vs. network? Many networks have multiple ASes.

needs to be destined for a core router and that by explicitly 

djs> needs to be destined towards a infrastructor router and that by
explicitly

   permitting only that known and understood traffic, the core devices

djs> the infrastructor devices 

   are not subjected to unnecessary traffic that might result in a
   denial of service attack.

djs> denial of service. The word attack implies your only addressing
actual intentional attacks.

   Since edge infrastructure access control lists protect only the
   infrastructure, the development of the list differs somewhat from
   "traditional" access filter lists:

          1. Review addressing scheme, and identify address block(s)
   that represent core devices.

djs> that represent infrastructor devices.

          2. Determine what traffic must be destined to the core devices
   from outside the AS.

djs> towards the infrastructor devices from outside the service
provider's network.

          3. Create a filter that allow the required traffic, denies all
   traffic destined to the core address block and then finally, permits
   all other traffic to all.

   As with other ingress filtering techniques, EIACLs are applied on
   ingress into the network, and clearly comprehensive coverage (i.e. on
   as many interface as possible) yields the most protection.

djs> EIACLs are applied as close to the ingress point as feasible 
djs> and the more comprehensive the coverage the better the protection.



3.2.  Other Traffic


   In addition to the explicitly permitted traffic, EIACLs can be
   combined with other common edge filters such as:

          1. Source spoof prevention (as per RFC 3704) by denying
   internal AS addresses as external sources.



Gill, Lewis, Quinn, Schoenmaker                   Section 3.2.  [Page 7]

INTERNET-DRAFT                  Expires:                       October
2006


          2. Filtering of reserved addresses (e.g. rfc1918 addresses) as
   traffic should not be sourced from reserved address.
          3. Other unneeded or unnecessary traffic

   Filtering this traffic can be part of the list explicitly or
   implicitly, however, explicit filters often provides log-able
   information that can be of use during a security event.




3.3.  Edge Infrastructure Conclusion


   Edge Infrastructure Access Control Lists provide a very effective
   first line of defense.  To deploy them effectively, core address
   space must be identifiable and widespread deployment is necessary.




4.  Edge Rewrite/Remarking


   RFC 1812 section 5.3 defines the use of IP Preference in IPv4 packets
   for routing, management and control traffic.  In addition it
   recommends devices use a mechanism for providing preferential
   forwarding for packets marked as routing, management or control
   traffic using IP Preference bits 6 or 7 (110 or 111 in binary.)
   RFC2474 defines DSCP and the compatibility of IP Preference bits when
   using DSCP.

   All packets received from the Customer edge (CE,) and the Peer Edge
   by the Provider Edge (PE,) with IP Preference values of 6 or 7 or
   DSCP bits of 11xxxx, as specified in RFC2474 Differentiated Services
   Field Definition, should have the IP Preference bits rewritten.
   Routing traffic received from the CE and the Peer Edge can safely
   have the IP Preference bits rewritten, because only a limited number
   of protocols are transmitted beyond the first PE router.  The bits
   may be rewritten to any value other than IP Preference values 6 or 7,
   or any DSCP value other than 11xxxx.  The new value can be based on
   the network operators IP Preference or DSCP policy.  If no policy
   exists the bits should be rewritten to 0.

djs> EBGP could be multihop and you may want to leave the DSCP bits
alone?

   Providers may not want to modify traffic that goes through their
   network in an effort offer a fully transparent service.  If the
   provider relies on alternative means of classifying traffic for
   prioritized forwarding rewriting the IP Preference bits is not



<SNIP>


4.1.  Edge Rewriting/Remarking Discussion


   By default router vendors do not differentiate an interface on a PE
   router connected to a P router from an interface connected to a CE
   router.  As a result any packet with the proper IP Preference or DSCP
   bits set may receive the same preferential forwarding behavior as
   legitimate routing, management, and control traffic.  A malicious
   attack may be able to take advantage of the vulnerability to increase
   the effectiveness of the attack or to attack the routing, management,
   and/or control traffic directly.

   This document is aimed at protecting network infrastructure from
   traffic to the device rather than traffic through the device.  Even
   though the edge rewrite/remarking deals primarily with traffic
   through a device it is included because the traffic has a direct
   impact on traffic to a device.  The forwarding prioritization given
   to routing, management, and control traffic by default leaves devices
   vulnerable to indirect attacks to the core infrastructure.

djs> vulnerable to indirect attacks to the infrastructure. 


<SNIP>


5.1.  Service Specific Access Control


   Typically these mechanisms are not concerned with protecting the
   system as a whole, but the service from exploitation.  The goal is
   not overall system availability, but maximizing the security of the
   particular service.

djs> The systems availablity is dependant on the security of the
services running on the system.
 
<SNIP>




5.2.3.  Access Control Implementation Guide


   Implementing a complex set of access controls for all traffic going
   to and from a router is non trivial.  The following is a recommended
   set of steps that has been used successfully by many carriers.

          -Develop list of required protocols
          -Develop source address requirements
          -Determine destination interface on router
          Does the protocol access a single interface?
          Does the protocol access many interfaces?
          Does the protocol access a virtual or physical interfaces?
          -Deployment should be an iterative process

djs>      -start with log instead of drop if possible and review the
logs.

          -Start with relatively open lists then tighten as needed



5.3.  Device Access Authorization and Accounting


   Operators should use per command authorization and accounting
   wherever possible.  Aside from their utility in mitigating other
   security threats, they provide an invaluable tool in the post event
   forensics.

djs> AAA seems out of place here. I agree its important and that logs of
commands 
djs> is a great after actions item to have.

<SNIP>


6.  Infrastructure Hiding


   Hiding the infrastructure of the network provides an elegant
   mechanism for protecting the network infrastructure.  If the
   destination of an attack is to an infrastructure address that is
   unreachable, attacks become far more difficult.  Infrastructure
djs> untargetable or difficult to target. 
djs> Its still really reachable via the transit traffic plane.
   hiding can be achieved in several ways:

<SNIP>

6.4.2.  Address core out of rfc1918 space


   In addition to filtering the visibility of core addresses to the
   wider Internet, it may be possible to use rfc1918 netblocks for
   numbering infrastructure when IP addresses are required (eg,
   loopbacks).  This added level of obscurity takes prevention of wide
   distribution of your infrastructure address space one step further.
   Many networks filter out packets with rfc1918 address at
   ingress/egress points as a matter of course.  In this circumstance,
   tools such as traceroute can work through your core, but reverse-
   resolution of descriptive names should be restricted to queries from
   internal/support groups.

djs> I think changing the default ports for services should be included
here.
djs> This is really service hiding but it can be used to limit your
exposure.

6.5 Change the ports services run on.

   Many attack tools, worms, auto-rooters hardcode in the destination
port.
   Changing that port to something that is not the well known port can 
   provide additional protection from such tools. This will not protect
you against
   a determined attacker. 

<SNIP>
Security through obscurity WORKS against some worms and other tools:)
Donald.Smith@qwest.com giac


This communication is the property of Qwest and may contain confidential or
privileged information. Unauthorized use of this communication is strictly 
prohibited and may be unlawful.  If you have received this communication 
in error, please immediately notify the sender by reply e-mail and destroy 
all copies of the communication and any attachments.