ToolDNS filtering limits
DNS filtering bypass test checklist
Use this checklist to prove your DNS filtering policy holds against hard-coded resolvers, DNS over HTTPS, DNS over TLS, VPNs, IPv6 and direct IP access on every network and VLAN. You can size the audit in test runs and hours, score a pass rate per network-test pair, and fix findings in order of exposure.
Watch the explainer
Use this to prove your DNS filtering policy holds against hard-coded resolvers, DNS over HTTPS (RFC 8484), VPNs and IPv6. At six networks, 12 tests and three device types, it returns 216 test runs and 18 hours of effort. Two failed tests per network give 12 open findings and an 83.3% pass rate.
What does a DNS filtering bypass test checklist tell you?
The checklist gives you three outputs for each network or virtual LAN (VLAN). You get a pass or fail against 12 bypass tests, a list of open findings, and a pass rate you can report. The planner beside it sizes the audit in test runs and hours before you book the time.
Start with placement, because placement decides what can go wrong. DNS filtering happens at the resolver. Your network hands each device a resolver address through DHCP, the service that assigns addresses when a device joins. Every lookup that reaches that resolver meets your policy.
With Purple Shield, that policy is set per network and per VLAN. Guest WiFi, Staff WiFi and an IoT VLAN can each carry different rules on the same access points.
Every bypass is a lookup that never reaches the resolver. A device can carry its own resolver address or wrap DNS in HTTPS or TLS. It can tunnel through a VPN, or skip DNS entirely by connecting straight to an IP address. None of these is a fault in the filter. They are gaps in the network around it, and only client-side testing shows them.
The DNS filtering limits pillar explains why each gap exists. This guide is the test that finds them on your estate.
A device's lookups pass through the DHCP-assigned resolver and the Shield VLAN policy, while hard-coded resolvers, encrypted DNS, IPv6 and direct IP routes bypass it unless gateway rules close them.
Connections
- Device on a network or VLAN to Resolver assigned by DHCP (standard lookup)
- Resolver assigned by DHCP to Shield policy for this VLAN
- Shield policy for this VLAN to Allowed or blocked
- Device on a network or VLAN to Hard-coded resolver (bypass)
- Device on a network or VLAN to Encrypted DNS in an app (bypass)
- Device on a network or VLAN to IPv6 resolver or direct IP (bypass)
- Hard-coded resolver to Gateway firewall and DHCP scope
- Encrypted DNS in an app to Gateway firewall and DHCP scope
- IPv6 resolver or direct IP to Gateway firewall and DHCP scope
The tests apply to any protective DNS service. Cisco Umbrella, Cloudflare Gateway and Purple Shield all enforce policy at the resolver. A network-level bypass therefore affects each of them the same way.
Umbrella and Cloudflare Gateway also offer device clients for managed laptops. Guest, shopper and fan devices are unmanaged, though, so you cannot install anything on them. On a venue network, the controls that close these gaps sit on your gateway and your DHCP scope.
How do you test a DNS filter for bypass routes?
Run the checklist on every network in scope, on each device type you support. Use one device per platform. Join it to the network the way a guest or staff member would, so it lands on the same VLAN they do.
Record each result as pass or fail. Then enter your counts into the planner to size the work and score the outcome.
DNS filtering bypass test checklist
0 of 12 done
Reading the four groups
Policy is live. These four tests prove the right policy reaches this network. If test one fails, stop and fix the VLAN mapping first. Every later result on that network is unreliable until you do.
Resolver bypass. These tests check the routes around the resolver.
- Port 53 is plain DNS.
- Port 853 is DNS over TLS (DoT), defined in RFC 7858.
- DNS over HTTPS (DoH) is defined in RFC 8484. It looks like ordinary web traffic on port 443.
Firefox checks a canary domain before it switches DoH on by default. A manual setting overrides that check, so test both cases. For Private Relay, follow Apple's network guidance.
Network path bypass. IPv6, direct IP and VPN tests check paths that skip the resolver altogether. DNS filtering cannot see a connection made straight to an IP address. The pass state there is a firewall rule, or a limit you have formally accepted and recorded.
Resilience. Test 12 checks the fallback. Most DHCP scopes list two resolvers, and the second one matters as much as the first.
DNS filtering bypass test planner
Enter your estate to size the audit and score the result. Each run is one test, on one device type, on one network.
Test runs
216
Testing effort
18 hours
Open findings
12
Policy pass rate
83.3 %
Worked example: At six networks, 12 tests, three device types, five minutes a run and two failures per network, the planner returns 216 test runs, 18 hours, 12 open findings and an 83.3% pass rate.
Assumptions
- A test fails for a network if it fails on any one device type.
- Each run includes joining the network, running the test and logging the result.
- Failures are averaged across networks, so real results can be uneven.
- Remediation time is not included. Budget it once you know the findings.
- Mobile data never touches the venue network and is not scored.
Worked example: a 200-room hotel with six networks
Here are the inputs, with every assumption stated.
- The venue. A 200-room hotel runs six networks on HPE Aruba access points: Guest WiFi, Staff WiFi, Conference WiFi, back office and POS, IoT and building systems, and an events network.
- The tester. One network administrator runs all 12 tests.
- The devices. Each test runs on three device types: a Windows laptop, an iPhone and an Android phone.
- The timing. Each run takes five minutes, including joining the network and logging the result.
The planner multiplies six networks by 12 tests by three devices to give 216 runs. At five minutes each, that is 1,080 minutes, or 18 hours. The administrator books two and a half working days.
The results came back uneven, as real estates do.
| Network or VLAN | Shield policy | Tests passed | Failed tests | First fix |
|---|---|---|---|---|
| Guest WiFi | Threats plus adult and gambling categories | 10 of 12 | DoH in browser; IPv6 resolver advertised | Align IPv6 resolver; block known DoH endpoints |
| Staff WiFi | Threats plus adult categories | 11 of 12 | Hard-coded resolver on port 53 | Block outbound port 53 except to the filtered resolver |
| Conference WiFi | Threats plus adult categories | 9 of 12 | DoT on port 853; DoH; VPN app | Block port 853 and DoH endpoints; review VPN rules |
| Back office and POS | Threats only, with payment allow-list | 12 of 12 | None | Retest after the next change |
| IoT and building systems | Threats only | 10 of 12 | Hard-coded device resolvers; direct IP calls | Firewall rules for port 53 and known destinations |
| Events network | None assigned at test time | 8 of 12 | Threat block, category block, public secondary resolver, Private Relay | Assign a VLAN policy; remove the public secondary |
Across six networks, 60 of 72 network-test pairs passed. The 12 failures average two per network, which matches the planner default. The pass rate is 60 divided by 72, or 83.3%.
The pattern tells you more than the score does. The events network was created for a conference and never assigned a policy. That one change accounts for a third of the findings. Conference WiFi left encrypted DNS open because its firewall rules were copied from an older template.
The back office passed everything. Its firewall allows outbound traffic only to named payment and management destinations.
How the planner builds its numbers
Each output follows from the inputs with no hidden weighting.
| Output | Formula | Worked example |
|---|---|---|
| Test runs | networks × tests × device types | 6 × 12 × 3 = 216 |
| Testing effort | test runs × minutes ÷ 60, rounded up | 1,080 ÷ 60 = 18 hours |
| Open findings | networks × average failed tests | 6 × 2 = 12 |
| Pass rate | (tests minus failed) ÷ tests × 100 | 10 ÷ 12 × 100 = 83.3% |
The pass rate counts network-test pairs, not device runs. A test that fails on one phone out of three still fails for that network. Device runs drive effort, and network-test pairs drive the score.
The planner caps failures at the number of tests, so the pass rate never drops below zero. It rounds effort up to the next whole hour, because a half-finished network gives you nothing to act on.
Two assumptions deserve a second look.
- The five-minute run. It assumes the tester already has the device joined and a test domain list ready. A first audit often runs slower, so raise the minutes input to seven or eight for your first pass.
- Mobile data. It is out of scope. A phone that drops to 4G or 5G leaves your network entirely, and no venue control reaches it. Mobile data and WiFi filtering explains how to word that limit in your acceptable use policy.
How to adapt the bypass test to your estate
Change the inputs
Count networks as the VLANs a person can land on, not the SSIDs (network names) they see. One SSID that assigns guests and staff to different VLANs counts as two networks.
On a large estate, full coverage grows fast. A retailer with 40 sites and two VLANs each enters 80 networks. At the default 12 tests, three devices and five minutes, the planner returns 2,880 runs and 240 hours.
The practical answer is a template approach:
- Run the full checklist on one site per network template.
- Run tests one, two and 12 everywhere else. They catch a missing policy or a public fallback resolver within minutes.
| Estate | Networks to enter | Device types | Tests to add or drop |
|---|---|---|---|
| Single hotel | Every VLAN a guest or staff member can join | 3 | Keep all 12 |
| Multi-site retail | Template sites in full, plus sampled sites | 2 to 3 | Drop the browser DoH test on POS-only VLANs; keep 1, 2 and 12 everywhere |
| Build-to-rent or student halls | Shared and amenity networks for residents | 4, adding a smart TV | Repeat the port 53 test per device class |
| Stadium | Concourse, hospitality and operations VLANs | 3 | Rerun before each season and major event |
| Healthcare | Patient, visitor and staff networks | 3 | Add clinical systems to test four |
Change the tests
Drop tests that cannot apply. An IoT VLAN with no browsers does not need the DoH browser test. It does need the port 53 test, because many devices ship with hard-coded resolvers.
Add tests your vertical demands. In healthcare, extend test four to cover clinical and booking systems, so a category rule never blocks patient care.
Change the cadence
Rerun the full checklist after any change that touches the path to the resolver:
- a new SSID or VLAN
- a gateway firmware update
- switching on IPv6
- a change of resolver policy
The events network in the worked example shows the cost of skipping this step.
What to do with your filtering audit result
Fix findings in order of exposure, not in checklist order.
- Policy not live. Failures on tests one to three mean a network is open. Assign the policy the same day.
- Resolver bypass on guest-facing networks. Block outbound ports 53 and 853 except to your filtered resolver, then block known DoH endpoints.
- IPv6. Make the IPv6 resolver match the IPv4 one, or the policy covers half the traffic. IPv6 and DNS filtering gaps walks through router advertisements and DHCPv6.
- Direct IP and VPN. Close these with firewall rules where you can, and record the rest as accepted limits. Direct IP access and DNS filtering explains why a blocked site can still load.
- False positives. A failure on test four costs you staff time and guest complaints. Use DNS filtering false positives to diagnose the cause before you allow-list anything.
Then rerun only the failed tests and update the pass rate. The worked example hotel closed its events network and Conference WiFi findings first. That recovered seven of 12 failures in an afternoon and lifted the pass rate to 93.1%, or 67 of 72.
Shield makes the policy side of this quicker. You set a policy per network and VLAN, standalone or on Purple Access, on the access points you already own. Shield fails closed, so an outage never leaves a network unfiltered.
See how Shield works for the policy model. Read the enterprise WiFi security guide for the wider controls around it. Shield is priced on a tailored quote, with a 30-day free trial to run this checklist before and after.
To adapt the tool, change the network count first, then the device types, then the test list. Your next read is IPv6 and DNS filtering gaps, the finding most estates miss on the first pass.
Key terms
- Protective DNS
- A recursive resolver that applies threat and category policy to DNS queries, defined by the DNS protocol in RFC 1035. It returns a block response or block page instead of the real address for disallowed domains. Cisco Umbrella, Cloudflare Gateway and Purple Shield all enforce policy at the resolver, so a network-level bypass affects each of them the same way and the same 12 tests apply.
- DHCP
- Dynamic Host Configuration Protocol (RFC 2131 for IPv4, RFC 8415 for DHCPv6), which assigns addresses and options, including primary and secondary DNS resolver addresses, when a device joins a network. The DHCP scope decides which resolver a device uses. Test one confirms the filtered resolver was handed out, and test 12 checks the secondary is not a public resolver.
- VLAN
- Virtual LAN, a logical layer 2 segment defined by IEEE 802.1Q tagging, letting one physical network carry several isolated broadcast domains. Shield sets policy per network and VLAN. Count networks as the VLANs a person can land on, not the SSIDs they see, when entering estate size into the planner.
- SSID
- Service Set Identifier, the network name broadcast by an access point under IEEE 802.11. One SSID can assign different users to different VLANs. Join each SSID as a guest or staff member would and test from the VLAN it lands on, because one SSID can map to two networks for audit purposes.
- DNS over TLS (DoT)
- DNS queries carried inside a TLS session on TCP port 853, specified in RFC 7858 to provide confidentiality between client and resolver. If port 853 is open at the gateway, a device can reach an unfiltered resolver over encrypted DNS. Test six checks that the port is blocked except to your filtered resolver.
- DNS over HTTPS (DoH)
- DNS queries sent as HTTPS requests on port 443, specified in RFC 8484, which makes them look like ordinary web traffic. Browser DoH can skip the network resolver entirely. The checklist tests it on default and manual browser settings and the fix is blocking known DoH endpoints.
- Canary domain
- The domain use-application-dns.net, which Firefox queries before enabling DoH by default. A network that returns NXDOMAIN or no answer signals that DoH should stay off. A manual DoH setting overrides the canary check, so you test both the default and the manual case to see whether your network relies on the signal alone.
- iCloud Private Relay
- An Apple service that routes Safari traffic and DNS through two relays. Apple documents how a network can signal that relay is unavailable by blocking or failing resolution of its relay hostnames. Test eight turns on Private Relay on an iPhone and loads a blocked domain. In the worked example, the events network failed this test.
- IPv6 resolver advertisement
- The DNS server address a device learns on IPv6 through router advertisements (RFC 8106 RDNSS option) or DHCPv6 (RFC 8415), separate from the IPv4 DHCP resolver. An unfiltered IPv6 resolver undoes the IPv4 policy. Switching on IPv6 is a change that should trigger a full rerun of the checklist.
- Fail closed
- A design where a security control blocks rather than allows traffic when it cannot reach its policy service, so an outage does not remove protection. Shield fails closed, but a public secondary resolver in DHCP can reopen the gap that fail-closed design shuts, which is why test 12 exists.
Worked examples
A 200-room hotel runs six networks on HPE Aruba access points: Guest WiFi, Staff WiFi, Conference WiFi, back office and POS, IoT and building systems, and an events network. How should one administrator size and act on a bypass audit?
Six networks by 12 tests by three device types (Windows laptop, iPhone, Android phone) gives 216 runs. At five minutes each that is 1,080 minutes, or 18 hours, so the administrator booked two and a half working days. Results showed 60 of 72 network-test pairs passed, an 83.3% pass rate. The events network had never been assigned a policy and accounted for a third of the findings. Conference WiFi left encrypted DNS open because its firewall rules were copied from an older template. Fixing those two networks first recovered seven of 12 failures in an afternoon, lifting the pass rate to 93.1%, or 67 of 72. The back office passed everything because its firewall allows outbound traffic only to named payment and management destinations.
A retailer with 40 sites and two VLANs each wants full bypass coverage. Entering 80 networks into the planner returns 2,880 runs and 240 hours. How do they keep the audit practical?
They adopt a template approach. The full 12-test checklist runs on one site per network template, which proves the configuration that every matching site inherits. Every other site runs only tests one, two and 12: confirming the device received the filtered resolver from DHCP, confirming a threat test domain is blocked, and confirming the secondary resolver is filtered rather than public. These three tests catch a missing policy or a public fallback resolver within minutes. On POS-only VLANs they drop the browser DoH test, because those networks carry no browsers, and use two to three device types rather than adding more.
Frequently asked questions
Do I need new hardware to run these tests or fix what they find?
No. The tests need a laptop, two phones and a command line. Every fix lands on equipment you already run: the DHCP scope, the gateway firewall and the resolver policy. Shield is protective DNS, so you point each network's resolver at it rather than installing an appliance. You can run it standalone or on Purple Access, with a separate policy for each network and VLAN.
Will the checklist work with the access points we already own?
Yes. DNS filtering sits above the access layer, so the checklist and Shield work the same on Cisco Meraki, HPE Aruba, Ruckus, Juniper Mist, Ubiquiti UniFi, Cambium, Extreme and Fortinet. What changes between vendors is where you set the DHCP resolver and the firewall rules. Follow your vendor's documentation for those settings, then run the same 12 tests.
How is Purple Shield priced?
Shield is priced on a tailored quote for your estate, and you can start with a 30-day free trial. The quote reflects the networks and VLANs you want covered. It also reflects whether you run Shield standalone or on Purple Access. Run this checklist during the trial to get a before-and-after pass rate on your own networks.
How does testing Shield differ from testing Cisco Umbrella or Cloudflare Gateway?
The tests are identical, because Cisco Umbrella, Cloudflare Gateway and Purple Shield all enforce policy at the resolver. A hard-coded resolver or an open port 853 bypasses any of them in the same way. The difference lies in what you configure afterwards. Shield sets policy per network and VLAN and fails closed, so a single outage never opens the network.
Can DNS filtering block mobile data, direct IP access or VPNs?
Not on its own. DNS filtering sees lookups, so a connection made straight to an IP address or tunnelled through a VPN needs a gateway firewall rule. Mobile data never touches your network, so no venue control reaches it. The checklist records each of these as a finding or an accepted limit. That way, nobody assumes coverage that does not exist.
How often should we rerun the bypass test?
Rerun the full checklist after any network change and at least once a quarter. Changes that matter include a new SSID or VLAN, a gateway firmware update, switching on IPv6 and a change of resolver policy. Between full runs, repeat tests one, two and 12 monthly on every network. They catch a missing policy or a public fallback resolver in minutes.
Does a filtering audit help with PCI DSS or ISO 27001?
Yes, as evidence. A dated log of tests, results and fixes shows an auditor that your filtering controls are checked, not assumed. That supports ISO 27001 control reviews and sits alongside the evidence a PCI DSS assessment asks for on payment networks. Purple holds ISO 27001 and Cyber Essentials certification and handles data in line with GDPR.
What happens to filtering if Shield is unreachable during a test?
Filtering holds. Shield fails closed, so an outage never leaves a network unfiltered. That is why test 12 matters. A public secondary resolver in your DHCP scope can reopen the gap that fail-closed design shuts. Remove the public secondary, then rerun tests two and 12 to confirm lookups stay on the filtered path.
Sources
- RFC 8484: DNS Queries over HTTPS (DoH)
- RFC 7858: DNS over Transport Layer Security (TLS)
- RFC 1035: Domain names, implementation and specification
- Mozilla: Canary domain use-application-dns.net
- Apple: Prepare your network for iCloud Private Relay
- ISO/IEC 27001 information security management
- PCI Security Standards Council: PCI DSS
