Executive brief
CoreDNS, a widely used DNS server and the default for Kubernetes clusters, contains a flaw that allows an attacker to crash the service. By sending a small number of specifically crafted network requests, an attacker can trigger a fatal error that shuts down the DNS server. This can lead to a complete loss of network connectivity and service discovery within a cloud environment, potentially causing widespread application failures.
Technical details
A denial of service (DoS) vulnerability exists in the CoreDNS 'loop' plugin due to the use of a predictable pseudo-random number generator (PRNG) and an insecure error handling mechanism. The plugin generates a secret query name (qname) using Go's math/rand package seeded with a predictable timestamp (UnixNano). If an attacker can predict or observe this qname (e.g., via logs during a network degradation window), they can send three matching HINFO queries. This triggers a fatal error handler (log.Fatalf) which calls os.Exit(1), immediately terminating the CoreDNS process. The vulnerability is most exploitable during the 30-second self-test window when upstream servers are unreachable. The issue is fixed in CoreDNS version 1.14.2 by switching to crypto/rand for stronger randomness.
Affected products
- CoreDNS CoreDNS < 1.14.2
- Red Hat Red Hat Advanced Cluster Management for Kubernetes 2.14 affected
- Red Hat Red Hat Advanced Cluster Management for Kubernetes 2.15 affected
- Red Hat Red Hat Connectivity Link 1 affected
Timeline
- 2026-03-06: disclosed
- 2026-03-06: advisory
- 2026-03-06: patched: Fixed in CoreDNS v1.14.2
References
- https://github.com/coredns/coredns/releases/tag/v1.14.2
- https://github.com/coredns/coredns/security/advisories/GHSA-h75p-j8xm-m278
- https://access.redhat.com/errata/RHSA-2026:25127
- https://access.redhat.com/errata/RHSA-2026:8151
- https://access.redhat.com/security/cve/CVE-2026-26018
- https://bugzilla.redhat.com/show_bug.cgi?id=2445242
- https://security.access.redhat.com/data/csaf/v2/vex/2026/cve-2026-26018.json