English
Securing the Domain Name System
The Domain Name System provides a critical service in the Internet infrastructure since it maps the domain names that are used by end users onto IP addresses. Since end users rely on names to identify the servers that they connect to, any incorrect information distributed by the DNS would direct end users' connections to invalid destinations. Unfortunately, several attacks of this kind occurred in the past. A detailed analysis of the security threats against the DNS appeared in :rfc:`3833`. We consider three of these threats in this section and leave the others to :rfc:`3833`.
The first type of attack is `eavesdropping`. An attacker who can capture packets sent to a DNS resolver or a DNS server can gain valuable information about the DNS names that are used by a given end user. If the attacker can capture all the packets sent to a DNS resolver, he/she can collect a lot of meta data about the domain names used by the end user. Preventing this type of attack has not been an objective of the initial design of the DNS. There are currently discussions with the IETF to carry DNS messages over TLS sessions to protect against such attacks. However, these solutions are not yet widely deployed.
The second type of attack is the `man-in-the-middle` attack. Consider that Alice is sending DNS requests to her DNS resolver. Unfortunately, Mallory sits in front of this resolver and can capture and modify all the packets sent by Alice to her resolver. In this case, Mallory can easily modify the DNS responses sent by the resolver to redirect Alice's packets to a different IP address controlled by Mallory. This enables Mallory to observe (and possibly modify) all the packets sent and received by Alice. In practice, executing this attack is not simple since DNS resolvers are usually installed in protected datacenters. However, if Mallory controls the WiFi access point that Alice uses to access the Internet, he could easily modify the packets on this access point and some software packages automate this type of attacks.
If Mallory cannot control a router on the path between Alice and her resolver, she could still launch a different attack. To understand this attack, it is important to correctly understand how the DNS protocol operates and the roles of the different fields of the DNS header which is reproduced in the figure below.
DNS header
The first field of the header is the `Identification` field. When Alice sends a DNS request, she places a 16-bits integer in this field and remembers it. When she receives a response, she uses this `Identification` field to locate the initial DNS request that she sent. The response is only used if its `Identification` matches a pending DNS request (containing the same question).
Mallory has studied the DNS protocol and understands how it works. If he can predict a popular domain for which Alice will regularly send DNS requests, then he can prepare a set of DNS responses that map the name requested by Alice to an IP address controlled by Mallory instead of the legitimate DNS response. Each DNS response has a different `Identification`. Since there are only 65,536 values for the `Identification` field, it is possible for Mallory to send them to Alice hoping that one of them will be received while Alice is waiting for a DNS response with the same identifier. In the past, it was difficult to send 65,536 DNS responses quickly enough. However, with the high speed links that are available today, this is not an issue anymore. A second concern for Mallory is that he must be able to send the DNS responses as if they were coming directly from the DNS resolver. This implies that Mallory must be able to send IP packets that appear to originate from a different address. Although networks should be configured to prevent this type of attack, this is not always the case and there are networks where it is possible for a host to send packets with a different source IP address [#fspoof]_. If the attack targets a single end user, e.g. Alice, this is annoying for this user. However, if the attacker can target a DNS resolver that serves an entire company or an entire ISP, the impact of the attack can be much larger in particular if the injected DNS response carries a long TTL and thus resides in the resolver's cache for a long period of time.
Fortunately, designers of DNS servers and resolvers have found solutions to mitigate this type of attack. The easiest approach would have been to update the format of the DNS requests and responses to include a larger `Identifier` field. Unfortunately, this elegant solution was not possible with the DNS because the DNS messages do not include any version number that would have enabled such a change. Since the DNS messages are exchanged inside UDP segments, the DNS developers found an alternate solution to counter this attack. There are two ways for the DNS library used by Alice to send her DNS requests. A first solution is to bind one UDP source port and always send the DNS requests from this source port (the destination port is always port ``53``). The advantage of this solution is that Alice's DNS library can easily receive the DNS responses by listening to her chosen port. Unfortunately, once the attacker has found the source port used by Alice, he only needs to send 65,536 DNS responses to inject an invalid response. Fortunately, Alice can send her DNS requests in a different way. Instead of using the same source port for all DNS requests, she can use a different source port for each request. In practice, each DNS request will be sent from a different source port. From an implementation viewpoint, this implies that Alice's DNS library will need to listen to one different port number for each pending DNS request. This increases the complexity of her implementation. From a security viewpoint there is a clear benefit since the attacker needs to guess both the 16 bits `Identifier` and the 16 bits `UDP source port` to inject a fake DNS response. To generate all possible DNS responses, the attacker would need to generate almost :math:`2^{32}` different messages, which is excessive in today's networks. Most DNS implementations use this second approach to prevent these cache poisoning attacks.
These attacks affect the DNS messages that are exchanged between a client and its resolver or between a resolver and name servers. Another type of attack exploits the possibility of providing several resource records inside one DNS response. A frequent optimization used by DNS servers and resolvers is to include several related resource records in each response. For example, if a client sends a DNS query for an `NS` record, it usually receives in the response both the queried record, i.e. the name of the DNS server that serves the queried domain, and the IP addresses of this server. Some DNS servers return several `NS` records and the associated IP addresses. The `cache poisoning` attack exploits this DNS optimization.
Let us illustrate it on an example. Assume that Alice frequently uses the `example.net` domain and in particular the web server whose name is `www.example.net`. Mallory would like to redirect the TCP connections established by Alice towards `www.example.net` to one IP address that he controls. Assume that Mallory controls the `mallory.net` domain. Mallory can tune the DNS server of his domain and add special DNS records to the responses that it sends. An attack could go roughly as follows. Mallory forces Alice to visit the `www.mallory.net` web site. He can achieve this by sending a spam message to Alice or buying advertisements on a web site visited by Alice and redirect one of these advertisements to `www.mallory.net`. When visiting the advertisement, Alice's DNS resolver will send a DNS request for `www.mallory.net`. Since Mallory control the DNS server, he can easily add in the response a `AAAA` record that associates `www.example.net` to the IP address controlled by Mallory. If Alice's DNS library does not check the returned response, the cache entry for `www.example.net` will be replaced by the `AAAA` record sent by Mallory.
To cope with these security threats and improve the security of the DNS, the IETF has defined several extensions that are known as DNSSEC. DNSSEC exploits public-key cryptography to authenticate the content of the DNS records that are sent by DNS servers and resolvers. DNSEC is defined in three main documents :rfc:`4033`, :rfc:`4034`, :rfc:`4035`. With DNSSEC, each DNS zone uses one public-private key pair. This key pair is only used to sign and authenticate DNS records. The DNS records are not encrypted and DNSSEC does not provide any confidentiality. Other DNS extensions are being developed to ensure the confidentiality of the information exchanged between a client and its resolvers :rfc:`7626`. Some of these extensions exchange DNS records over a TLS session which provides the required confidentiality, but they are not yet deployed and outside the scope of this chapter.
DNSSEC defines four new types of DNS records that are used together to authenticate the information distributed by the DNS.
the `DNSKEY` record allows storing the public key associated with a zone. This record is encoded as a TLV and includes a `Base64` representation of the key and the identification of the public key algorithm. This allows the `DNSKEY` record to support different public key algorithms.
the `RRSIG` record is used to encode the signature of a DNS record. This record contains several sub-fields. The most important ones are the algorithm used to generate the signature, the identifier of the public key used to sign the record, the original TTL of the signed record and the validity period for the signature.
the `DS` record contains a hash of a public key. It is used by a parent zone to certify the public key used by one of its child zones.
the `NSEC` record is used when non-existent domain names are queried. Its usage will be explained later
The simplest way to understand the operation of DNSSEC is to rely on a simple example. Let us consider the `example.org` domain and assume that Alice wants to retrieve the `AAAA` record for `www.example.org` using DNSSEC.
The security of DNSSEC relies on `anchored keys`. An `anchored key` is a public key that is considered as trusted by a resolver. In our example, we assume that Alice's resolver has obtained the public key of the servers that manage the root zone in a secure way. This key has been distributed outside of the DNS, e.g. it has been published in a newspaper or has been received in a sealed letter.
To obtain an authenticated record for `www.example.org`, Alice's resolver first needs to retrieve the `NS` which is responsible for the `.org` Top-Level Domain (TLD). This record is served by the DNS root server and Alice's resolver can retrieve the signature (`RRSIG` record) for this `NS` record. Since Alice knows the `DNSKEY` of the root, she can verify the validity of this signature.