Executive brief
containerd is a core container runtime used by Kubernetes and container orchestration systems to manage containerized workloads. A resource exhaustion bug in its CRI plugin can cause the containerd daemon to crash due to memory depletion when containers run exec probes or lifecycle hooks with long-lived background processes. An attacker or misconfigured container can trigger this crash repeatedly, disrupting container operations across an entire node until the service is manually restarted.
Technical details
The vulnerability exists in containerd's CRI ExecSync implementation, which handles container execution probes (liveness, readiness checks) and lifecycle hooks. When these operations launch background child processes, the I/O drain phase lacks timeout or context cancellation handling, causing stdio-drain goroutines to block indefinitely. Repeated ExecSync calls accumulate these blocked goroutines, leaking memory and goroutine resources over time. An attacker with the ability to deploy containers or influence probe configurations can exploit this via specially crafted containers or probes. The vulnerability affects Linux systems running containerd with CRI plugin enabled; the issue does not affect Windows or non-CRI deployments. Patches are available in containerd versions 1.7.35, 2.0.12, 2.2.8, and 2.3.5.
Affected products
- containerd containerd < 1.7.35
- containerd containerd 2.0.x before 2.0.12
- containerd containerd 2.2.x before 2.2.8
- containerd containerd 2.3.x before 2.3.5
Timeline
- 2026-09-09: disclosed: Vulnerability disclosed in GitHub Security Advisory GHSA-7jxh-36q5-gcqv
- 2026-09: patched: Patches released: containerd 1.7.35, 2.0.12, 2.2.8, 2.3.5