Skip to content
chornous.dev

Type two or more letters. Esc closes.

Index of sheets
Theme

Sheet 03 · Writing

All notes

Java4Shell: CVE-2026-114938 lets crafted Base64 attack the HotSpot JIT in Java 17 through 27

OpenJDK and Oracle shipped the November 17, 2026 Java security update with ten OpenJDK CVEs. CVE-2026-114938, nicknamed Java4Shell, is a memory corruption bug in the x86-64 HotSpot Base64 decoder that an unauthenticated attacker can reach over the network in Java 17, 21, 25 and 27.

Pl. 55 · section drawing generated from the slug “java4shell-cve-2026-114938-openjdk”

The OpenJDK Vulnerability Group published its November 17, 2026 advisory with ten CVEs, and Oracle shipped the matching fixes the same day in its November Critical Security Patch Update. The one with a name is CVE-2026-114938. CISA calls it Java4Shell and describes it as memory corruption in the Base64 decoder of the x86-64 HotSpot JVM. Its alert says crafted Base64 input can crash the JVM and may allow unauthenticated remote code execution.

Oracle scores it 8.1 under CVSS 3.1 and files it under the HotSpot compiler. Albert Yang and Quang Luong of Calif.io reported it.

Affected builds

The OpenJDK advisory lists the affected builds as 27.0.1.0.3, 25.0.5.0.3, 21.0.13.0.3, 17.0.21.0.3, 11.0.33, 8u512 and earlier. For CVE-2026-114938 the risk matrix marks 17, 21, 25 and 27. It leaves 8 and 11 unmarked.

Oracle's own list adds two product lines you might forget about:

Oracle products affected by CVE-2026-114938
ProductAffected versions
Oracle Java SE8u511b33-perf, 17.0.21.0.3, 21.0.13.0.3, 25.0.5.0.3, 27.0.1.0.3
Oracle GraalVM for JDK17.0.21.0.3, 21.0.13.0.3
Oracle GraalVM Enterprise Edition21.3.20.0.3

Note the 8u511b33-perf entry. If you run the Java 8 performance pack, you're in scope even though the OpenJDK matrix skips 8.

Oracle rates the attack complexity as high and lists no privileges and no user interaction. Its advisory says a successful attack "can result in takeover" of Java SE. Oracle also spells out the path: an attacker supplies data to APIs in the affected component, for example through a web service. Applets and Web Start don't play a part. A service that decodes Base64 from a request body, a header or a JWT on an affected JVM fits that description.

The rest of the update

The OpenJDK matrix lists nine more CVEs, from 114939 to 114947, and Oracle counts 11 Java SE patches once you add the OpenJFX libxml2 fix. Oracle says an attacker can exploit all 11 over the network without authentication.

Higher-scored CVEs in the November 17 OpenJDK advisory
CVEComponentCVSS 3.1Versions
CVE-2026-114938hotspot/compiler8.117, 21, 25, 27
CVE-2026-114939security-libs/org.ietf.jgss7.48 through 27
CVE-2026-114940security-libs/javax.net.ssl7.48 through 27
CVE-2026-114941hotspot/compiler7.021, 25, 27
CVE-2026-114942core-libs7.017, 21, 25, 27
CVE-2026-86144javafx/web (OpenJFX)7.521, 25, 27

Oracle credits Nicholas Carlini of Anthropic for the JSSE bug, CVE-2026-114940. Researchers from Paderborn University and Ruhr-University Bochum found CVE-2026-114947, a lower-scored JSSE issue in 25 and 27.

Find every JVM first

CISA's first recommendation is an inventory, and it points at the hard part: JDKs bundled inside third-party applications, container images and appliances often don't show up as their own line item. Start with your images:

# Every java binary and its version, per image
for img in $(docker images --format '{{.Repository}}:{{.Tag}}'); do
  echo "== $img"
  docker run --rm --entrypoint sh "$img" -c \
    'for j in $(find / -type f -name java -perm -u+x 2>/dev/null); do "$j" -version 2>&1 | head -1; done' \
    2>/dev/null
done

Then look inside the vendor tools on your servers: CI runners, search clusters, build agents, IDE installs on shared hosts. Each one might ship its own runtime.

If you can't patch today

CISA lists compensating controls: validate and restrict Base64 input at the application boundary, enforce input-size limits, and limit unauthenticated access. In a Spring or Jakarta app, a size cap in front of the decoder takes a few lines:

private static final int MAX_B64 = 16 * 1024;

static byte[] decodeBounded(String input) {
    if (input == null || input.length() > MAX_B64) {
        throw new IllegalArgumentException("base64 payload too large");
    }
    return Base64.getDecoder().decode(input);
}

Treat that as a stopgap. CISA names patching as the fix, and a size cap won't protect a library that decodes Base64 inside code you don't own.

CISA also asks you to watch for unexpected JVM crashes and unknown Java processes. A spike of hs_err_pid*.log files on an internet-facing host deserves a look this week.

This week

  1. Pull your vendor's November 17 JDK build for 17, 21, 25 and 27 and roll it to internet-facing services first. Oracle's next scheduled update lands on December 15, 2026.
  2. Include GraalVM and the Java 8 performance pack in the sweep.
  3. Rebuild container images on the new base and redeploy. Patching the host does nothing for a JDK baked into an image.
  4. If a rebuild waits on a vendor, cap Base64 input size at the edge and alert on JVM crashes.

Volodymyr Chornous