Guides
Labs: DNS enumeration without touching real infrastructure
These exercises teach subdomain enumeration on fictional zones. -simulate-zone
loads a scenario file and answers every query from a DNS server that subenum
runs on 127.0.0.1 for the length of the scan. No DNS traffic leaves your
machine, so labs are safe in a classroom, a CTF or an air-gapped VM.
Unlike -simulate, which invents results from a hit rate, a lab runs the real
resolver against the scenario. Wildcard detection, -rate pacing, retries, the
reliability guard and the run-quality report all behave exactly as they would
against a live zone, and the results are the same on every run.
Setup
You need a subenum binary (see Installation).
The scenarios and their wordlist are built into it, so the labs run from any
directory: -simulate-zone lab1 (or lab1-first-scan) picks the built-in
scenario, and without -w the scan uses the labs’ short wordlist. For a
zero-install classroom, open the labs in GitHub Codespaces:
the container builds ./subenum, opens this page and the first scenario, and
prints the lab 1 command.
Every lab scans lab.example. To edit a scenario or write your own, copy one
from examples/labs
and pass its path instead, for example -simulate-zone ./my-lab.zone; a path
uses the bundled 5,000-word list unless you add -w.
Labs run from the command line. Combining -tui with an active -simulate-zone
setting (including SUBENUM_SIMULATE_ZONE or a config-file value) exits with
usage error 2 before the interface opens. Remove -tui to run the lab.
Each lab ends with questions. The answers are folded underneath; try first.
Lab 1: a first scan
subenum -simulate-zone lab1 -show-records lab.example
-show-records prints the records behind each name.
- How many subdomains did the scan find?
- Which host has an IPv6 address?
shophas the same address aswww. Why?- The scenario file defines
blog, but it is not in the results. Rerun with-type A,AAAA,CNAME. What isblog, and why did the default scan miss it?
Answers
- Five:
www,mail,vpn,jenkinsandshop. www(2001:db8::10).shopis a CNAME (an alias) forwww. The resolver follows the alias and returnswww’s addresses. The CNAME run showsCNAME=www.lab.example.blogis a CNAME toghost-lab.example.net, a host that does not exist. The default lookups (A and AAAA) find no address, so the name looks empty. With CNAME lookups, subenum shows the alias and flags itTAKEOVER?=dangling: the alias points at a name nobody has registered. If that were a cloud service, whoever claimed the name would controlblog.lab.example. This is how subdomain takeovers start. See takeover hints in the README.
Lab 2: spot the wildcard
subenum -simulate-zone lab2 lab.example
- What does subenum report, and why does it refuse to continue?
- Rerun with
-force. Which names are reported now, and how does subenum tell them apart from the wildcard?
Answers
- “Wildcard DNS detected”: the zone has a
*record, so every name, even a random one, resolves to192.0.2.250. Without filtering, every word in the list would be reported as a “finding”. www,portalandintranet. Before scanning, subenum queries several random names to fingerprint the wildcard’s answer. With-forceit then drops every result that matches that fingerprint; these three have their own addresses, so they are real hosts. Thewildcard-filteredline counts the rest.
Lab 3: rate limits
This resolver answers REFUSED once it gets more than 40 queries in a second, as many public and corporate resolvers do.
subenum -simulate-zone lab3 -stats lab3.json lab.example
subenum -simulate-zone lab3 -stats lab3-paced.json -rate 30 lab.example
- Compare the
refusedcounts of the two runs. - Open both
-statsreports. What is each run’sverdict? - Why can the first run still list some real names, and why should you not trust it anyway?
Answers
- The first run bursts with 100 workers and most of its lookups are refused.
The paced run (
-rate 30, under the resolver’s 40-per-second limit) has none. unreliablefor the burst,completefor the paced run.- The first queries arrived before the limit kicked in, so whatever they found is real. But a REFUSED name was never actually checked: it might exist. A run with many refusals says nothing about the names it missed, which is why the report calls it unreliable. Slower is faster here.
Lab 4: recursive discovery
subenum -simulate-zone lab4 lab.example
subenum -simulate-zone lab4 -recursive -depth 3 lab.example
- How many names does the flat scan find? How many does the recursive scan find?
- Which name needed
-depth 3? - Try the recursive scan with the bundled 5,000-word list (add
-w ""). What happens, and why?
Answers
- Three (
www,dev,corp) against eight. Recursion takes every name it finds and enumerates the wordlist under it, which turns upapi.dev,git.dev,staging.devandvpn.corp, thenadmin.staging.dev. admin.staging.dev, three labels belowlab.example.- subenum refuses to start: with 5,000 words, depth 3 could mean over 100
billion queries. Recursion multiplies the wordlist at every level, so it
asks you to set
-max-queries(or-force) first. Against a real target, that guard is what keeps a scan from becoming a flood. A shorter list for the deeper levels also keeps it small:-recursive-wnames the labels tried under each name found, so-wwith 5,000 words and-recursive-wwith 100 cost 5,000 queries for depth 1 and 100 for each name found.
Lab 5: when lookups fail
subenum -simulate-zone lab5 -timeout 300 -stats lab5.json lab.example
- Which outcome lines in the breakdown are non-zero besides
resolvedandnxdomain? - What is the run’s
verdictinlab5.json, and what does it mean for the names that failed?
Answers
timeout(legacyandoldnever answer),refused(beta) andother(backupanswers SERVFAIL).degraded: the scan finished, but some names could not be checked. A failed lookup is not “does not exist”. Those names need a retry (-attempts) or a different resolver before you can rule them out.
Lab 6: IPv6-only services
subenum -simulate-zone lab6 -show-records lab.example
subenum -simulate-zone lab6 -type A -show-records lab.example
subenum -simulate-zone lab6 -type AAAA -show-records lab.example
- Which three names does the default scan find, and which address families does each have?
- Which name disappears with
-type A? Does that mean the service is absent? - Which name disappears with
-type AAAA? Why does the default scan find both?
Answers
wwwhas bothA=192.0.2.10andAAAA=2001:db8::10;apihas onlyAAAA=2001:db8::20;legacyhas onlyA=192.0.2.30.apidisappears. It exists but has no IPv4 address: an A-only lookup cannot discover an IPv6-only service.legacydisappears because it has no IPv6 address. The default types areA,AAAA, and a record of either type is enough to report a name. Querying both families does not require the machine to have IPv6 connectivity: this lab’s DNS server still answers on IPv4 loopback (127.0.0.1).
Lab 7: a zone transferunreleased
subenum -simulate-zone lab7 -format jsonl lab.example
subenum -simulate-zone lab7 -axfr -format jsonl lab.example
The lab’s name server hands its whole zone to anyone who asks, the
misconfiguration -axfr looks for. In lab mode the transfer goes to the lab
server, never to the name server’s real address.
- Which names does the second run find that the first does not, and what is
their
sources? - What does the second run print on stderr about the name server?
- Copy
examples/labs/lab7-zone-transfer.zone, delete its$TRANSFER allowline and run the copy with-axfr -v. What changes?
Answers
billing-legacy,k8s-ingress-eu2andns1: no wordlist holds them. Theirsourcesis["axfr"], whilewww,mailandvpnsay["wordlist", "axfr"]. Every transferred name is still looked up, so a stale entry would not be reported.- That
ns1.lab.exampleallowed a transfer oflab.exampleand that anyone can list the zone: a finding to report in itself.-statsnames the server underaxfr_allowed. - The server refuses, as most do. The scan is the wordlist’s alone, and
-vsays that no name server allowed a transfer.
Lab 8: an NSEC walkunreleased
subenum -simulate-zone lab8 -format jsonl lab.example
subenum -simulate-zone lab8 -nsec-walk -format jsonl lab.example
The zone is signed with plain DNSSEC NSEC and refuses transfers. Every
answer that says a name does not exist also names the next name that does,
so -nsec-walk can list the whole zone, one query per name. In lab mode
the walk goes to the lab server.
- Which names does the second run find that the first does not, and what is
their
sources? - How many queries did the walk take, and why that many?
- Copy
examples/labs/lab8-nsec-walk.zone, change$DNSSEC nsecto$DNSSEC nsec3and run the copy with-nsec-walk. What changes?
Answers
payroll-archive,qa.build-farmandns1: no wordlist holds them. Theirsourcesis["nsec"], whilewww,mailandvpnsay["wordlist", "nsec"]. Every walked name is still looked up.- Seven: one per name in the chain, the apex included, until the last
record leads back to the apex.
-statscounts the names undernsec_names. - NSEC3 records hold hashes of names, not names, so there is nothing to follow: subenum says it is not walking the zone and why, and the scan is the wordlist’s alone.
Lab 9: a dangling delegationunreleased
subenum -simulate-zone lab9 -format jsonl lab.example
subenum -simulate-zone lab9 -type A,NS -format jsonl -stats run.json lab.example
legacy is a subzone delegated to two name servers that no longer exist, so
the lab’s resolver, which cannot reach either, answers SERVFAIL for it, as a
real one would. Whoever registers legacy-dns.example would answer for every
name under legacy. staging is delegated to a name server that still
answers. With NS in -type, subenum asks the zone’s own name server about a
name that ends SERVFAIL, without asking it to recurse, and gets the
delegation back as a referral. In lab mode that query goes to the lab
server.
- How does
legacyend in the first run, and in the second? - What does the second run’s report say about the extra queries?
- Why is
stagingnot marked?
Answers
- The first run counts it under
other(inconclusive): withoutNSnothing asks the zone. The second reports it with its two NS records and"takeover_candidate":"dangling-ns", because both name servers are NXDOMAIN. parent_checksis 1 (one name was asked about) andparent_queriesis 1;queries_sentincludes it. The query takes a-rateslot like any other.- Its name server,
ns1.lab.example, exists, so a resolver gets its NS records directly and there is nothing to take over.
Writing your own scenario
A scenario is a plain text file. Names are relative to the domain you scan,
and # starts a comment:
$DELAY 20ms # every answer waits this long
$REFUSE-ABOVE 50 # answer REFUSED past 50 queries a second
$TRANSFER allow # answer a zone transfer (-axfr); the default is refuse
$DNSSEC nsec # prove absent names with NSEC (-nsec-walk), or nsec3; the default is none
@ A 192.0.2.1 # the scanned domain itself
www A 192.0.2.10
www AAAA 2001:db8::10
shop CNAME www # relative target
blog CNAME ghost.example.net. # absolute target (trailing dot)
*.dev A 192.0.2.99 # wildcard under dev
mail MX 10 mx1 # preference and host (relative or absolute)
@ TXT "v=spf1 -all" # the rest of the line; quotes optional
@ HTTPS 1 . alpn=h2,h3 port=443 # priority, host ("." or a name), alpn and port (-type HTTPS)
cdn HTTPS 0 edge.example.net. # AliasMode: points cdn at another host, like a CNAME
sub NS ns1.gone.example. # a delegated subzone (scan with -type NS)
slow TIMEOUT # never answered
broken SERVFAIL # also: REFUSED, NXDOMAIN
192.0.2.10 PTR host-12.dc1 # reverse DNS of an address (-ptr); relative or absolute host
Start a scenario in examples/labs with a comment saying what it teaches
(# Lab 6: <title>. <what to look for>): subenum mcp shows that comment,
without the lab number and any indented command line, to agents choosing a
scenario.
Names the file does not define are NXDOMAIN. The scenario is the whole world
during a lab, so names outside the scanned domain (such as an external CNAME
target) are NXDOMAIN too. A CNAME to a name in the scenario is followed as a
recursive resolver follows it, through further CNAMEs (up to 8) and
wildcards, so a CNAME b, b CNAME www answers with www’s addresses. An
AAAA value must be a real IPv6 address: an IPv4-mapped one such as
::ffff:192.0.2.1 is rejected. An answer too big for the client’s UDP
buffer comes back truncated, so the client asks again over TCP. Use the
documentation ranges (192.0.2.0/24, 198.51.100.0/24, 203.0.113.0/24,
2001:db8::/32) and reserved names such as lab.example, so nothing in a
lab points at a real host.
Scenarios work with the other flags, such as -dL (each listed domain gets
the same zone), -permute, -diff, -format json and -stats, which
records the scenario file under zone. With -axfr or -nsec-walk, the
transfer, the DNSSEC probe and the walk go to the lab server, whatever
address the scenario gives its name servers. With $DNSSEC nsec, an answer
for a name the file does not define carries the NSEC record that covers it,
from a chain of the apex, the defined names (not failing ones) and the
wildcards. A delegation none of whose name servers has an address in the
scenario (such as sub above) answers SERVFAIL for itself and every name
under it, as a resolver that cannot reach them does, and a query that does
not ask for recursion gets a referral to them, as from the parent zone’s
server; a delegation with one name server that answers is served from the
file. -simulate-zone cannot be combined
with -simulate, -dns-server, -r, -ct or -sources, whether they
come from the command line, a SUBENUM_* variable or the config file: a lab run refuses to start
rather than send a query off the machine.