ZDNS is a command-line utility that provides high-speed DNS lookups. ZDNS is written in Go and contains its own recursive resolution code and a cache optimized for performing lookups of a diverse set of names. We use https://github.com/zmap/dns to construct and parse raw DNS packets. For more information about ZDNS's architecture and performance, check out the following paper appearing at ACM's Internet Measurement Conference '22.
As an example, the following will perform MX lookups and a secondary A lookup for the IPs of MX servers for the domains in the Alexa Top Million:
cat top-1m.csv | ./zdns MX --ipv4-lookup --alexa
ZDNS can be installed by checking out the repository and running make zdns.
git clone https://github.com/zmap/zdns.git
cd zdns
make zdnsZDNS was originally built as a CLI tool only. Work has been done to convert this library into a standalone library and let a separate CLI wrap the library.
The library consists of a ResolverConfig struct which will contain all config options for all lookups made.
The ResolverConfig is used to create 1+ Resolver struct(s) which will make all lookups. A Resolver
should only make a single lookup at a time (it is not thread-safe) and multiple Resolver structs should be
used for parallelism. See our examples for how to use the
library. Modules are used to define the behavior of the lookups.
ZDNS provides several types of modules:
-
Raw DNS modules provide the raw DNS response from the server similar to dig, but in JSON. There is a module for (nearly) every type of DNS record
-
Lookup modules provide more helpful responses when multiple queries are required (e.g., completing additional
Alookup for IP addresses if aNSis received inNSLOOKUP) -
Misc modules provide other additional means of querying servers (e.g.,
bind.version)
We detail the modules below:
The A, AAAA, AFSDB, ANY, ATMA, AVC, AXFR, BINDVERSION, CAA, CDNSKEY, CDS, CERT, CNAME, CSYNC, DHCID, DMARC, DNSKEY, DS, EID, EUI48, EUI64, GID, GPOS, HINFO, HIP, HTTPS, ISDN, KEY, KX, L32, L64, LOC, LP, MB, MD, MF, MG, MR, MX, NAPTR, NID, NINFO, NS, NSAPPTR, NSEC, NSEC3, NSEC3PARAM, NSLOOKUP, NULL, NXT, OPENPGPKEY, PTR, PX, RP, RRSIG, RT, SVCBS, MIMEA, SOA, SPF, SRV, SSHFP, TALINK, TKEY, TLSA, TXT, UID, UINFO, UNSPEC, and URI modules provide the raw DNS response in JSON form, similar to dig.
For example, the command:
echo "censys.io" | ./zdns A
returns:
{
"name": "censys.io",
"results": {
"A": {
"data": {
"additionals": [
{
"flags": "",
"type": "EDNS0",
"udpsize": 512,
"version": 0
}
],
"answers": [
{
"answer": "104.18.10.85",
"class": "IN",
"name": "censys.io",
"ttl": 300,
"type": "A"
},
{
"answer": "104.18.11.85",
"class": "IN",
"name": "censys.io",
"ttl": 300,
"type": "A"
}
],
"protocol": "udp",
"resolver": "[2603:6013:9d00:3302::1]:53"
},
"duration": 0.285295416,
"status": "NOERROR",
"timestamp": "2024-08-23T13:12:43-04:00"
}
}
}Raw DNS responses frequently do not provide the data you want. For example,
an MX response may not include the associated A records in the additional
section requiring an additional lookup. To address this gap and provide a
friendlier interface, we also provide several lookup modules: alookup,
mxlookup, and nslookup.
alookup acts similar to nslookup and will follow CNAME records.
mxlookup will additionally do an A lookup for the IP addresses that correspond with an exchange record.
nslookup will additionally do an A/AAAA lookup for IP addresses that correspond with an NS record
For example,
echo "censys.io" | ./zdns mxlookup --ipv4-lookup
returns:
{
"name": "censys.io",
"results": {
"MXLOOKUP": {
"data": {
"exchanges": [
{
"class": "IN",
"ipv4_addresses": [
"209.85.202.27"
],
"name": "alt1.aspmx.l.google.com",
"preference": 5,
"ttl": 300,
"type": "MX"
},
{
"class": "IN",
"ipv4_addresses": [
"142.250.31.26"
],
"name": "aspmx.l.google.com",
"preference": 1,
"ttl": 300,
"type": "MX"
}
]
},
"duration": 0.154786958,
"status": "NOERROR",
"timestamp": "2024-08-23T13:10:11-04:00"
}
}
}ZDNS also supports special "debug" DNS queries. Modules include: BINDVERSION.
ZDNS supports providing input in a variety of formats depending on the desired behavior.
The most basic input is a list of names separated by newlines. For example:
From stdin:
echo "google.com\nyahoo.com" | ./zdns A
cat list_of_domains.txt | ./zdns A
From a file
./zdns A --input-file=list_of_domains.txtIf you don't need to resolve many domains, providing the domain as CLI argument, similar to dig, is supported for ease-of-use.
For example:
./zdns A google.com --name-servers=1.1.1.1Equivalent to dig -t A google.com @1.1.1.1
Normally, ZDNS will choose a random nameserver for each domain lookup from --name-servers. If instead you want to specify
a different name server for each domain, you can do so by providing domainName,nameServerIP pairs seperated by newlines.
This will override any nameservers provided with --name-servers.
For example:
echo "google.com,1.1.1.1\nfacebook.com,8.8.8.8" | ./zdns A
You can see the resolver is as specified for each domain in the output (additionals/answers redacted for brevity):
$ echo "google.com,1.1.1.1\nfacebook.com,8.8.8.8" | ./zdns A
{"name":"google.com","results":{"A":{"data":{"additionals":...,"answers":[...],"protocol":"udp","resolver":"1.1.1.1:53"},"duration":0.030490042,"status":"NOERROR","timestamp":"2024-09-13T09:51:34-04:00"}}}
{"name":"facebook.com","results":{"A":{"data":{"additionals":[...],"answers":[...],"protocol":"udp","resolver":"8.8.8.8:53"},"duration":0.061365459,"status":"NOERROR","timestamp":"2024-09-13T09:51:34-04:00"}}}ZDNS can either operate against a recursive resolver (e.g., an organizational DNS server) [default behavior] or can perform its own recursion internally. If you are performing a small number of lookups (i.e., millions) and using a less than 10,000 go routines, it is typically fastest to use one of the common recursive resolvers like Cloudflare or Google. Cloudflare is nearly always faster than Google. This is particularly true if you're looking up popular names because they're cached and can be answered in a single round trip. When using tens of thousands of concurrent threads, consider performing iteration internally in order to avoid DOS'ing and/or rate limiting your recursive resolver.
To perform local recursion, run zdns with the --iterative flag. When this
flag is used, ZDNS will round-robin between the published root servers (e.g.,
198.41.0.4). In iterative mode, you can control the size of the local cache by
specifying --cache-size and the timeout for individual iterations by setting
--iteration-timeout. The --timeout flag controls the timeout of the entire
resolution for a given input (i.e., the sum of all iterative steps).
ZDNS performance stems from massive parallelization using light-weight Go routines. This architecture has several caveats:
-
Every Go routine uses its own dedicated network socket. Thus, you need to be able to open as many sockets (in terms of both max file descriptors and ephemeral ports) as you have threads specified (via
--threads). By default, ZDNS uses 1,000 threads, which is less than Linux's default max number of 1024 open FDs. However, it is greater than Mac OS's default of 256. You can view the maximum number of open FDs (and thus sockets) permitted by runningulimit -n. If you want to run with a greater number of threads than this number, you need to increase the number of open files at the OS level. If you fail to do this, you'll encounter a fatal error similar toFATA[0000] unable to create socketlisten udp <client IP address>:0: socket: too many open files. If you want to run more threads than you have ephemeral ports available, you will need to use multiple client IP addresses:--local-addr=A,B,C. -
By default, ZDNS "reuses" UDP sockets by creating an unbound UDP socket for each light-weight routine at launch and using it for all queries (regardless of destination IP). This dramatically improves performance because ZDNS and the host OS don't need to setup and tear down a socket to send each individual packet (since DNS queries/responses tend to be one packet each). However, this means that ZDNS will preallocate a socket for each thread at launch. This may not be optimal if you're only looking up a small number of names. For example, if you only need to lookup 100 names, but use the default 1,000 threads, you'll bind but never use 900 UDP sockets. Instead, of worrying about recycling sockets, we recommend that you specify a reasonable number of threads for your use case (since this also foregoes any work to start those threads in the first place). This is why, though, you can get an error about being unable to open a large number of sockets even though you're only looking up a single name. If it's important to create a fresh socket for each query, you can disable this reuse by specifying
--recycle-sockets=false. -
Go is happy to use all CPU cores that are available to it, and can use a tremendous amount of CPU if you specify a large number of threads. CPU is primarily used for parsing and JSON encoding. If you want to limit the number of CPU cores, you can do so by including the
--go-processes=nflag or setting theGOMAXPROCSenvironment variable. -
It's difficult to recommend a precise amount of
--threadsas it depends on several factors. The graph below shows how a sample workflow has lower runtime but higher rates of name resolution failure as the number of threads increases.
Much of the performance that you'll see depends on your workflow, hardware, and how many name servers the load is spread out on. If you're looking to maximize performance
for your workflow/hardware, we recommend starting at 100 threads and increasing until you start to see an increase in name resolution failures.
To help with this, you can use --output-file=output.jsonl and grep -v "NOERROR" output.jsonl | wc -l to count the number of names that failed to resolve.
Flags that may be of use in tuning performance are:
--timeoutThe maximum amount of time ZDNS will spend on a single name--iteration-timeoutThe maximum amount of time ZDNS will spend on a single iteration step (ex: resolving google.com at the .com layer)--network-timeoutThe maximum amount of time ZDNS will wait for a response from a nameserver--retries=NIf a connection to a specific nameserver fails in--iterative, ZDNS will retry with another un-queried name server at that layer. Retries are per-name, so if--retries=1then ZDNS will retry a name against a new nameserver once during it's full iteration process. If all nameservers have been queried then a random nameserver will be chosen.--name-serversThe list of nameservers to use for lookups, mostly useful with--iterative=false
DNS includes a lot of extraneous data that is not always useful. There are four
result verbosity levels: short, normal (default), long, and trace:
short: Short is the most terse result output. It contains only information about the responsesnormal: Normal provides everything included in short as well as data about the responding serverlong: Long outputs everything the server included in the DNS packet, including flags.trace: Trace outputs everything from every step of the recursion process
Users can also include specific additional fields using the --include-fields
flag and specifying a list of fields, e.g., --include-fields=flags,resolver.
Additional fields are: class, protocol, ttl, resolver, flags, dnssec.
By default ZDNS expects to receive a list of names to lookup on a small number of name servers. For example:
echo "google.com" | ./zdns A --name-servers=8.8.8.8,8.8.4.4
However, there are times where you instead want to lookup the same name across a large number of servers. This can be accomplished using name server mode. For example:
echo "8.8.8.8" | ./zdns A --name-server-mode --override-name="google.com"
Here, every line piped in ZDNS is sent an A query for google.com. ZDNS also
supports mixing and matching both modes by piping in a comma-delimited list of
name,nameServer. For example:
echo "google.com,8.8.8.8" | ./zdns A will send an A query for
google.com to 8.8.8.8 regardless of what name servers are specified by
--name-servers= flag. Lines that do not explicitly specify a name server will
use the servers specified by the OS or --name-servers flag as would normally
happen.
There is a feature available to perform a certain DNS query against all nameservers. For example, you might want to get the A records from all nameservers of a certain domain. To do so, you can do:
echo "google.com" | ./zdns A --all-nameservers
ZDNS supports using multiple lookup modules in a single invocation. For example, let's say you want to perform an A,
AAAA, and MXLOOKUP for a set of domains and you want to perform them with iterative resolution. You will need to use the
MULTIPLE module and provide a config file with the modules and module-specific flags you want to use.
Please see ./zdns --help and ./zdns <MODULE_NAME> --help for Global and Module-specific options that can be used in the config file.
For example:
cat 1000k_domains.txt | ./zdns MULTIPLE --multi-config-file="./multiple.ini"
Where multiple.ini is a file that looks like:
; Specify Global Options here
[Application Options]
iterative=true
; List out modules and their respective module-specific options here. A module can only be listed once
[MXLOOKUP]
ipv4-lookup = true
; You can use default values and just list modules if you don't need to specify any options
[A]
[AAAA]
A sample multiple.ini file is provided in github.com/zmap/zdns/cli/multiple.ini
By default, ZDNS will operate with 1,000 light-weight go routines. If you're
not careful, this will overwhelm many upstream DNS providers. We suggest that
users coordinate with local network administrators before performing any scans.
You can control the number of concurrent connections with the --threads and
--go-processes command line arguments. Alternate name servers can be
specified with --name-servers. ZDNS will rotate through these servers when
making requests. We have successfully run ZDNS with tens of thousands of
light-weight routines.
If zdns encounters a record type it does not support it will generate an output
record with the type field set correctly and a representation of the
underlying data structure in the unparsed_rr field. Do not rely on the
presence or structure of this field. This field (and its existence) may change
at any time as we expand support for additional record types. If you find
yourself using this field, please consider submitting a pull-request adding
parser support.
There is a benchmark available in benchmark/ that can be used to run ZDNS in a predictable fashion and print out some statistics about the run.
This can be useful for comparing performance before and after a change to ZDNS.
See more details in the benchmark README.
If you're interested in contributing to ZDNS, see CONTRIBUTING.
- Please use Github issues for filing bugs.
ZDNS Copyright 2020 Regents of the University of Michigan
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See LICENSE for the specific language governing permissions and limitations under the License.