Detection rules › Elastic

Potential DNS Rebinding from Public to Private Address

Severity
medium
Time window
15m
Group by
Esql.client_ip, dns.question.name, dns.question.registered_domain
Author
Elastic
Source
github.com/elastic/detection-rules

Identifies a client resolving the same public registered domain to both a public IP address and a private, loopback, link-local, unique-local IPv6, or shared address. This includes both address classes being observed at the same timestamp, and a public answer followed within five minutes by a private answer where the minimum TTL across all answer records in the private-answer events is 60 seconds or less. Either pattern is consistent with DNS rebinding that pivots browser or application trust to internal resources.

Known false positives

  • Split-horizon DNS, VPN transitions, service discovery, failover, hairpin NAT, and dual-stack names that publish a public A record with a unique-local AAAA record can legitimately produce public and private answers for the same name. Recursive resolvers, DNS forwarders, and localhost listeners can also aggregate many endpoints under one client address. Security products may sinkhole suspicious domains to loopback or private addresses with short TTLs. Confirm the domain, resolver placement, and client identity before adding an exception, and scope exceptions by registered domain or client rather than globally.

MITRE ATT&CK coverage

TacticTechniques
Initial Access

Rule body

[metadata]
creation_date = "2026/08/20"
integration = ["network_traffic", "zeek"]
maturity = "production"
updated_date = "2026/08/25"

[rule]
author = ["Elastic"]
description = """
Identifies a client resolving the same public registered domain to both a public IP address and a private, loopback,
link-local, unique-local IPv6, or shared address. This includes both address classes being observed at the same
timestamp, and a public answer followed within five minutes by a private answer where the minimum TTL across all answer
records in the private-answer events is 60 seconds or less. Either pattern is consistent with DNS rebinding that pivots
browser or application trust to internal resources.
"""
false_positives = [
    """
    Split-horizon DNS, VPN transitions, service discovery, failover, hairpin NAT, and dual-stack names that publish a
    public A record with a unique-local AAAA record can legitimately produce public and private answers for the same
    name. Recursive resolvers, DNS forwarders, and localhost listeners can also aggregate many endpoints under one
    client address. Security products may sinkhole suspicious domains to loopback or private addresses with short TTLs.
    Confirm the domain, resolver placement, and client identity before adding an exception, and scope exceptions by
    registered domain or client rather than globally.
    """,
]
from = "now-15m"
language = "esql"
license = "Elastic License v2"
name = "Potential DNS Rebinding from Public to Private Address"
note = """## Triage and analysis

### Investigating Potential DNS Rebinding from Public to Private Address

DNS rebinding uses attacker-controlled public names that resolve to internal addresses so a victim browser or client
reaches RFC1918, loopback, link-local, unique-local IPv6, or shared-address targets. This rule alerts on two patterns
for the same client and fully qualified domain name: public and internal addresses whose first observations have the
same timestamp, or a public answer followed by an internal answer within five minutes. Sequential transitions also
require the minimum TTL across all DNS answer records in the private-answer events to be 60 seconds or less. Equal
timestamps often represent a mixed-answer response, but do not prove that the addresses came from one DNS transaction;
parallel A and AAAA events can share a timestamp. Equal-timestamp matches do not require the TTL or five-minute gates.

`Esql.client_ip` is `client.ip` when present and otherwise `source.ip`. Depending on sensor placement this value may
identify an endpoint, a recursive resolver, a forwarder, or a localhost DNS listener such as `127.0.0.1`. Resolver or
loopback identities can merge many hosts into one bucket.

### Possible investigation steps

- Review `dns.question.name`, `dns.question.registered_domain`, `Esql.public_ips`, and `Esql.private_ips` to confirm the
  same name resolved to both a public and an internal address.
- Check `Esql.same_timestamp`. A value of `true` means both address classes were first observed at the same timestamp,
  but does not establish that they came from one DNS transaction. Compare `Esql.first_public_answer`,
  `Esql.first_private_answer`, `Esql.transition_seconds`, and `Esql.min_private_event_ttl` for sequential transitions.
  Short transitions and TTL values near zero increase confidence.
- Use `Esql.resolved_ip_observation_count`, `Esql.resolved_ip_count`, `Esql.dataset_values`, and
  `Esql.observer_name_values` to assess observation volume, distinct IP cardinality, and the integrations and sensors
  that contributed to the alert.
- Identify the requesting client using `Esql.client_ip`. Confirm whether that address is an endpoint rather than a
  recursive resolver, forwarder, or localhost DNS service before attributing the activity to one host.
- Determine whether the registered domain is attacker-controlled or a legitimate split-horizon or failover domain.
- Check the requesting host for browser or application connections to the resolved private address immediately after
  the private answer.
- Review the targeted internal service for requests carrying the public domain in the HTTP Host header or TLS SNI and
  for unauthorized access, state changes, or sensitive-data retrieval.

### False positive analysis

- Split-horizon DNS, VPN transitions, service discovery, failover, and hairpin NAT can legitimately cause a public
  registered domain to alternate between public and private answers. Confirm the domain and resolver behavior with DNS
  administrators.
- Dual-stack names may publish a public IPv4 address and a unique-local IPv6 address under the same question name.
- Security products may intentionally sinkhole suspicious domains to loopback or private addresses with short TTLs.
- Exclude confirmed internal domains, approved sinkholes, and controlled security-testing infrastructure by registered
  domain or client only after validation. Do not exclude a resolver address until the originating endpoint is known.

### Response and remediation

- Block or sinkhole the queried name at recursive resolvers if it is confirmed malicious.
- Patch or isolate affected internal services reached through the rebound name.
- Restrict outbound DNS for clients that should not resolve arbitrary external names directly.
"""
references = ["https://owasp.org/www-community/attacks/DNS_Rebinding", "https://portswigger.net/web-security/ssrf"]
risk_score = 47
rule_id = "c485ceb7-b0e3-4540-9e8c-e9c655406e68"
setup = """## Setup

This rule requires DNS transaction events from one of the following passive network integrations:

- Elastic Network Packet Capture (`network_traffic.dns`) in `logs-network_traffic.dns-*`
- Zeek (`zeek.dns`) in `logs-zeek.dns-*`
- Legacy Packetbeat DNS events in `packetbeat-*`

Enable DNS logging so that `dns.question.registered_domain` and `dns.resolved_ip` are populated. Populate
`dns.answers.ttl` to detect sequential public-to-private transitions; equal-timestamp matches do not require TTL data.

Place the sensor where it observes endpoint-to-resolver DNS traffic. If the sensor is upstream of a recursive resolver,
or if the captured client is a localhost listener such as `127.0.0.1`, `Esql.client_ip` may identify shared DNS
infrastructure instead of the originating endpoint and can merge answers from many hosts.

DNS-over-HTTPS (DoH), DNS-over-TLS (DoT), and other encrypted DNS traffic are not visible to packet capture unless the
sensor receives decrypted DNS telemetry or equivalent resolver logs mapped to ECS.
"""
severity = "medium"
tags = [
    "Domain: Network",
    "Use Case: Threat Detection",
    "Use Case: Network Security Monitoring",
    "Tactic: Initial Access",
    "Rule Type: ESQL",
    "Data Source: Network Packet Capture",
    "Data Source: Network Traffic",
    "Data Source: Zeek",
    "Resources: Investigation Guide",
]
timestamp_override = "event.ingested"
type = "esql"

query = '''
from logs-network_traffic.dns-*, logs-zeek.dns-*, packetbeat-*
| where
    (
      data_stream.dataset in ("network_traffic.dns", "zeek.dns") or
      event.dataset == "dns"
    ) and
    dns.question.name is not null and
    dns.question.registered_domain is not null and
    dns.resolved_ip is not null and
    TO_UPPER(dns.response_code) == "NOERROR" and
    TO_UPPER(dns.question.type) in ("A", "AAAA")
| eval
    Esql.client_ip = COALESCE(client.ip, source.ip),
    Esql.dataset = COALESCE(data_stream.dataset, event.dataset)
| where Esql.client_ip is not null
| mv_expand dns.resolved_ip
| eval Esql.is_private = CIDR_MATCH(
    dns.resolved_ip,
    "0.0.0.0/32",
    "10.0.0.0/8",
    "100.64.0.0/10",
    "127.0.0.0/8",
    "169.254.0.0/16",
    "172.16.0.0/12",
    "192.168.0.0/16",
    "::1/128",
    "fc00::/7",
    "fe80::/10"
  )
| eval
    Esql.private_time = CASE(Esql.is_private, @timestamp, null),
    Esql.public_time = CASE(not Esql.is_private, @timestamp, null),
    Esql.private_event_ttl = CASE(Esql.is_private, MV_MIN(dns.answers.ttl), null),
    Esql.private_ip = CASE(Esql.is_private, dns.resolved_ip, null),
    Esql.public_ip = CASE(not Esql.is_private, dns.resolved_ip, null)
| stats
    Esql.resolved_ip_observation_count = COUNT(*),
    Esql.resolved_ip_count = COUNT_DISTINCT(dns.resolved_ip),
    Esql.first_public_answer = MIN(Esql.public_time),
    Esql.first_private_answer = MIN(Esql.private_time),
    Esql.min_private_event_ttl = MIN(Esql.private_event_ttl),
    Esql.public_ips = MV_SLICE(VALUES(Esql.public_ip), 0, 100),
    Esql.private_ips = MV_SLICE(VALUES(Esql.private_ip), 0, 100),
    Esql.dataset_values = MV_SLICE(VALUES(Esql.dataset), 0, 10),
    Esql.observer_name_values = MV_SLICE(VALUES(observer.name), 0, 20)
  by Esql.client_ip, dns.question.name, dns.question.registered_domain
| eval
    Esql.same_timestamp = Esql.first_public_answer == Esql.first_private_answer,
    Esql.transition_seconds = DATE_DIFF("seconds", Esql.first_public_answer, Esql.first_private_answer)
| where
    Esql.first_public_answer is not null and
    Esql.first_private_answer is not null and
    (
        Esql.same_timestamp or
        (
            Esql.first_public_answer < Esql.first_private_answer and
            Esql.min_private_event_ttl is not null and
            Esql.min_private_event_ttl <= 60 and
            Esql.transition_seconds <= 300
        )
    )
| keep Esql.*, dns.*
'''


[[rule.threat]]
framework = "MITRE ATT&CK"
[[rule.threat.technique]]
id = "T1189"
name = "Drive-by Compromise"
reference = "https://attack.mitre.org/techniques/T1189/"


[rule.threat.tactic]
id = "TA0001"
name = "Initial Access"
reference = "https://attack.mitre.org/tactics/TA0001/"

[rule.investigation_fields]
field_names = [
    "Esql.client_ip",
    "dns.question.name",
    "dns.question.registered_domain",
    "Esql.resolved_ip_observation_count",
    "Esql.resolved_ip_count",
    "Esql.first_public_answer",
    "Esql.first_private_answer",
    "Esql.transition_seconds",
    "Esql.same_timestamp",
    "Esql.min_private_event_ttl",
    "Esql.public_ips",
    "Esql.private_ips",
    "Esql.dataset_values",
    "Esql.observer_name_values",
]

Stages and Predicates

Stage 1: from

from logs-network_traffic.dns-*, logs-zeek.dns-*, packetbeat-*

Stage 2: where

| where
    (
      data_stream.dataset in ("network_traffic.dns", "zeek.dns") or
      event.dataset == "dns"
    ) and
    dns.question.name is not null and
    dns.question.registered_domain is not null and
    dns.resolved_ip is not null and
    TO_UPPER(dns.response_code) == "NOERROR" and
    TO_UPPER(dns.question.type) in ("A", "AAAA")

Stage 3: eval

| eval
    Esql.client_ip = COALESCE(client.ip, source.ip),
    Esql.dataset = COALESCE(data_stream.dataset, event.dataset)

Stage 4: where

| where Esql.client_ip is not null

Stage 5: mv_expand

| mv_expand dns.resolved_ip

Stage 6: eval

| eval Esql.is_private = CIDR_MATCH(
    dns.resolved_ip,
    "0.0.0.0/32",
    "10.0.0.0/8",
    "100.64.0.0/10",
    "127.0.0.0/8",
    "169.254.0.0/16",
    "172.16.0.0/12",
    "192.168.0.0/16",
    "::1/128",
    "fc00::/7",
    "fe80::/10"
  )

Stage 7: eval

| eval
    Esql.private_time = CASE(Esql.is_private, @timestamp, null),
    Esql.public_time = CASE(not Esql.is_private, @timestamp, null),
    Esql.private_event_ttl = CASE(Esql.is_private, MV_MIN(dns.answers.ttl), null),
    Esql.private_ip = CASE(Esql.is_private, dns.resolved_ip, null),
    Esql.public_ip = CASE(not Esql.is_private, dns.resolved_ip, null)
Esql.private_event_ttl =
ifEsql.is_privateMV_MIN(dns.answers.ttl)
elsenull
Esql.private_ip =
ifEsql.is_privatedns.resolved_ip
elsenull
Esql.private_time =
ifEsql.is_private@timestamp
elsenull
Esql.public_ip =
ifnot Esql.is_privatedns.resolved_ip
elsenull
Esql.public_time =
ifnot Esql.is_private@timestamp
elsenull

Stage 8: stats

| stats
    Esql.resolved_ip_observation_count = COUNT(*),
    Esql.resolved_ip_count = COUNT_DISTINCT(dns.resolved_ip),
    Esql.first_public_answer = MIN(Esql.public_time),
    Esql.first_private_answer = MIN(Esql.private_time),
    Esql.min_private_event_ttl = MIN(Esql.private_event_ttl),
    Esql.public_ips = MV_SLICE(VALUES(Esql.public_ip), 0, 100),
    Esql.private_ips = MV_SLICE(VALUES(Esql.private_ip), 0, 100),
    Esql.dataset_values = MV_SLICE(VALUES(Esql.dataset), 0, 10),
    Esql.observer_name_values = MV_SLICE(VALUES(observer.name), 0, 20)
  by Esql.client_ip, dns.question.name, dns.question.registered_domain

Stage 9: eval

| eval
    Esql.same_timestamp = Esql.first_public_answer == Esql.first_private_answer,
    Esql.transition_seconds = DATE_DIFF("seconds", Esql.first_public_answer, Esql.first_private_answer)

Stage 10: where

| where
    Esql.first_public_answer is not null and
    Esql.first_private_answer is not null and
    (
        Esql.same_timestamp or
        (
            Esql.first_public_answer < Esql.first_private_answer and
            Esql.min_private_event_ttl is not null and
            Esql.min_private_event_ttl <= 60 and
            Esql.transition_seconds <= 300
        )
    )

Stage 11: keep

| keep Esql.*, dns.*

Indicators

These rows show field, operator, and value matches.

Output fields

These fields are emitted when the rule matches.

FieldSource
Esql.*KEEP Esql.*
dns.*KEEP dns.*