Domain Name System

Domain Name System (DNS) #

An aside: “Hourglass Model of the Internet”

  1. Email/general Application layer
  2. DNS/TLS
  3. TCP/UDP - up till here, the internet we interact with
  4. IP/BGP (narrowest part, i.e., everything needs to go through here)
  5. Link layer, Ethernet, Cellular

If a layer’s security is compromised, everything above it could be compromised.

Motivation for DNS #

  • Communication on the internet is via IP, but it’s hard to remember IP addresses!
  • But: easier to remember names, so map names to IP addresses (e.g. ->
  • Originally, a centralized solution - a hosts.txt file that has a mapping for all hosts:
    • Organization: host -> IP address
    • Store in /etc/hosts on system
    • Historically, Stanford Research Institute (SRI) kept main copy, a single place to update records by downloading them periodically
    • However: fragile, hard to scale and keep in sync
  • Now, a decentralized solution:
    • Root nameserver has a mapping: organization -> name of organization nameserver
    • Organization nameserver has a mapping: host -> IP address

DNS hierarchical namespace #

  1. DNS root ., owned by ICANN (prev. US Dept of Commerce, now independent nonprofit)
  2. TLDs; e.g. com., edu., us., held by TLD authoritative nameservers (e.g. Educause, Verisign), authority delegated from ICANN
  3. Domain: e.g., Stanford authoritative nameservers (authority delegated from Educause to Stanford)
  4. Subdomain: e.g., authority delegated from domain owner

DNS resolution #

  1. Client host: has Client Stub Resolver, asks recursive resolver to resolve a domain (e.g.
  2. Recursive resolver only has location for Root Authoritative NS, queries for
    • Root NS: responds to ask .edu authoritative NS, with its location
  3. Recursive resolver asks .edu auth NS, queries for
    • .edu auth NS: responds to ask and, with their locations
  4. Recursive resolver asks for
    • responds
  5. Recursive resolver responds to client stub resolver

Note: recursive resolvers will not tell you the IP addresses of nameservers, but authoritative nameservers will

DNS query structure #

Queries happen over UDP, with the following fields:

  1. UDP header
  2. DNS header
    • ID
    • QR
    • Opcode
    • AA
    • TC
    • RD
    • RA
    • Z
    • Rcode
      • 0: NOERROR (success)
      • 2: SERVFAIL (server failed to complete request)
      • 3: NXDOMAIN (domain name does not exist)
      • 4: NOTIMP (function not implemented)
      • 5: REFUSED (server refused to answer queries)
    • Query count
    • Answer count
    • NS count
    • Additional record count
  3. Question section
    • Query name (QNAME): domain to resolve
    • Query class (QCLASS)
      • IN: internet
      • CHAOS: used for debugging
    • Query type (QTYPE)
      • NS: authoritative nameserver for domain
      • A: IPv4 address
      • AAAA: IPv6 address
      • MX: mail exchange records
  4. Answer section
    • Name
    • Type
    • Class
    • TTL (how long to cache answer)
    • RDLength
    • RDATA
  5. Authority section
  6. Additional info section

Caching #

  • Why cache DNS responses?
    • Reduces load
    • Improves latency
    • Reuse results of previous queries
  • How long to cache? Governed by Time to Live (TTL)

DNS failures #

  • Misconfigurations: typos, misconfigured auth NS
  • Hardware/network failures: unreachable nameservers
  • Large traffic volumes; DoS attacks

DNS reliability #

  • UDP used, because faster than TCP (no need to do the whole handshake for a short single-use query)
  • Reliability through replication
    • Two auth NS per domain
    • Multiple root NS’s, TLD authoritative NS’s

DNS confidentiality #

  • Everyone on path can see your DNS queries!
  • DNS-over-TLS/DNS-over-HTTPS solves this, by directly routing queries to DNS providers like Google and Cloudflare
    • This made ISPs livid, because they were selling data about DNS queries they could collect in the clear
    • However, ISPs can now try to pretend to be - used for censorship

DNS administration #

  • Before 1999: US Department of Commerce, SRI-NIC, Network Solutions
  • After 1999: Commerce delegated to ICANN:
    • gTLD (generic TLD) have to obey rules set forth by ICANN
    • ccTLD (country code TLD) rules are determined by each country
  • Principles:
    • Competition
    • Representation
    • Stability
    • Private, bottom-up coordination
  • Stakeholders:
    • ICANN
    • Registrar (e.g. GoDaddy): handles registration of domains (i.e., selling domains)
    • Registry (e.g. Verisign): handles operations
    • Registrant (e.g. Google): domain owner

Updating nameservers #

  1. Registrant uses Registrar’s web API to change nameservers
  2. Registrar updates EPP (Extensible Provisioning Protocol) server held by Registry, which updates its database
  3. Registry updates TLD authoritative NS

Attacks on DNS infrastructure #

  • Domain hijacking: compromise of domain registrars to redirect a victim company’s domain name to a malicious IP
    • Used for conducting phishing attacks
  • TLS doesn’t help here!
    • TLS protects against MITM attacks
    • Automated TLS certificate issuance using domain validation (e.g. Let’s Encrypt) uses DNS to authenticate domain “ownership”
    • However, if attacker controls DNS on registrar end, can obtain TLS certificate for the fraudulent domain
    • 2017 attack against aerospace companies was pre-Cert Transparency, so no record of the new fraudulent certificate being issued
    • But: if you’re someone big like Google, then you ship your cert to browsers and have the browsers SSL pin you, so this attack is hard to pull off against such targets