Skip to main content
Vulnerability Database/CVE-2026-95838

CVE-2026-95838: containerd DOS Vulnerability

CVE-2026-95838 is a denial of service vulnerability in containerd that allows malicious images to trigger repeated directory traversals, consuming CPU and delaying container startup. This article covers technical details, affected versions, impact assessment, and mitigation strategies.

Published:

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Sources: National Vulnerability Database, CISA KEV, Exploit-DB

CVE-2026-95838 Overview

CVE-2026-95838 is a resource exhaustion vulnerability in containerd, an open-source container runtime widely used by Kubernetes and Docker deployments. The flaw resides in the layer unpack handler, which repeatedly traverses directories when processing certain opaque whiteout configurations. A crafted image layer can trigger these redundant directory walks during image pull or import operations. The condition is tracked under CWE-400: Uncontrolled Resource Consumption. containerd maintainers fixed the issue in versions 1.7.35, 2.0.12, 2.2.8, and 2.3.5.

Critical Impact

Pulling or importing a malicious image consumes excess CPU, delays container startup, and reduces image-pull concurrency for unrelated workloads on the same host.

Affected Products

  • containerd versions prior to 1.7.35
  • containerd 2.x versions prior to 2.0.12, 2.2.8, and 2.3.5
  • Downstream platforms embedding vulnerable containerd releases (Kubernetes nodes, Docker engines)

Discovery Timeline

  • 2026-10-12 - CVE-2026-95838 published to NVD
  • 2026-10-16 - Last updated in NVD database

Technical Details for CVE-2026-95838

Vulnerability Analysis

The vulnerability affects containerd's tar archive unpack logic in pkg/archive/tar.go. During layer extraction, containerd handles opaque whiteout markers, which signal that a directory in a lower layer should appear empty in the merged filesystem view. The unpack handler walks the target directory each time it encounters an opaque whiteout, even if the same directory has already been processed.

A malicious image can embed multiple opaque whiteout entries referencing the same directory tree. Each entry forces another full traversal, scaling CPU cost linearly with the number of markers. The condition does not corrupt data or escape the container boundary, but it degrades host performance and starves concurrent pull operations.

Root Cause

The unpack handler tracked individual unpacked paths but did not track directories that had already been subjected to an opaque whiteout walk. Without deduplication, the handler repeated identical filesystem walks for every opaque marker.

Attack Vector

Exploitation requires an operator to pull or import a crafted image. The attack vector is local to the host performing the unpack, and user action (the pull or import request) is required. The impact is availability degradation on the container host and queued image operations.

go
// Patch excerpt from pkg/archive/tar.go
// archive: skip redundant opaque whiteout walks
		// Used for handling opaque directory markers which
		// may occur out of order
		unpackedPaths = make(map[string]struct{})
+		// Used for tracking directories that have already processed an
+		// opaque directory marker to avoid redundant directory walks
+		opaqueDirs = make(map[string]struct{})

		convertWhiteout = options.ConvertWhiteout

Source: containerd commit e6ca378. The patch introduces an opaqueDirs map that records directories already walked, allowing subsequent opaque whiteout entries for the same path to be skipped.

Detection Methods for CVE-2026-95838

Indicators of Compromise

  • Sustained high CPU usage attributed to the containerd process during image pull or import operations.
  • Image pull latency that scales abnormally with layer count rather than layer size.
  • Backlog of pending container creations on nodes that have recently pulled a specific image.

Detection Strategies

  • Monitor containerd ImagePull and ImageImport gRPC call durations against baseline distributions and alert on statistical outliers.
  • Inspect image layer manifests for excessive numbers of opaque whiteout entries (.wh..wh..opq) targeting the same directory paths.
  • Correlate node-level CPU saturation events with concurrent containerd image operations in the container runtime logs.

Monitoring Recommendations

  • Enable containerd structured logging and ship events to a centralized SIEM for correlation with host performance metrics.
  • Track Kubernetes ImagePullBackOff and ContainerCreating state durations per node to surface runtime slowdowns.
  • Audit registries and image provenance to detect images from untrusted sources entering production pipelines.

How to Mitigate CVE-2026-95838

Immediate Actions Required

  • Upgrade containerd to 1.7.35, 2.0.12, 2.2.8, or 2.3.5 or later on all nodes.
  • Restrict image pulls to trusted registries and enforce image signing with policies such as Sigstore or Notation.
  • Review Kubernetes imagePullPolicy and admission controls to block unverified images from reaching worker nodes.

Patch Information

The fix is published in GitHub Security Advisory GHSA-rp3h-jf77-q9p4 and merged in containerd commit e6ca378. Operators running managed Kubernetes services should confirm their provider has rolled out containerd releases containing the fix.

Workarounds

  • Limit image pulls to signed, scanned images originating from internal registries while patches are deployed.
  • Apply resource limits (cpu.max, cpuset) to the containerd cgroup slice to bound the blast radius of a malicious unpack operation.
  • Pre-pull and cache approved images on nodes to reduce exposure to on-demand pulls from external sources.
bash
# Verify installed containerd version on each node
containerd --version

# Example upgrade on a Debian/Ubuntu host (adjust for your distribution)
sudo apt-get update
sudo apt-get install --only-upgrade containerd.io
sudo systemctl restart containerd

# Confirm the patched version is active
ctr version

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Sources: National Vulnerability Database, CISA KEV, Exploit-DB

Experience the Most Advanced Cybersecurity Platform

See how the world’s most intelligent, autonomous cybersecurity platform can protect your organization today and into the future.