· via dev.to (home feed)
Java4Shell patches land, but apt upgrades leave container and bundled JREs exposed
Patches for the HotSpot Base64 heap overflow shipped on November 17, but OS package upgrades do not reach the JVMs inside containers, vendor appliances and bundled runtimes.

The patch and the illusion of coverage
Fixes for Java4Shell, a heap overflow in HotSpot's AVX-512 Base64 decoder tracked as CVE-2026-114938, shipped on November 17, 2026. Calif, the researcher who found it, rates the bug 9.2 under CVSS 4.0 and lists Keycloak, Jira, Confluence, Artifactory, Jenkins and PingFederate as open to pre-auth remote code execution in default configurations. Oracle scores it 8.1 under CVSS 3.1, citing high attack complexity, according to Aikido.
The practical problem, as a dev.to walkthrough stresses, is that the reflexive fix — upgrading the system's openjdk package and confirming with java -version — repairs exactly one runtime. Containers ship their own JREs, CI agent images are often pinned to older Temurin builds, and vendor appliances carry JVMs that no package manager tracks. A current host binary says nothing about any of them.
What is actually affected
Per the dev.to post, exploitation requires all of the following: an affected Java 17, 21, 25 or 27 build (Calif dates the exposure from 17.0.6); an x86-64 CPU with AVX-512 VBMI, meaning Intel Xeon Ice Lake or newer, or AMD Zen 4 and above; a JVM not already constrained by -XX:UseAVX=2 or -XX:-UseBASE64Intrinsics; and attacker-controlled Base64 reaching java.util.Base64.Decoder. ARM64 is out of scope, and older EPYC generations (Naples, Rome and Milan) lack the required instructions.
Aikido says standard Java 8 and 11 are unaffected, except that Oracle lists its Java SE 8 Enterprise Performance Pack build 8u511-perf as vulnerable. In Aikido's roundup, the fixed Oracle builds are 17.0.21.1, 21.0.13.1, 25.0.5.1 and 27.0.1.1, with Temurin and Corretto equivalents carrying their own suffixes. End-of-life releases (18 through 20, 22 through 24, and 26) get no dedicated fix and must move to a supported line. Red Hat is expediting RHEL 8, 9 and 10 builds.
Reachable input is common: Harness points to HTTP Basic headers, JWT segments decoded before signature checks, and SASL exchanges such as Kafka SCRAM.
Exploitability is contested
The dev.to author calls it the worst JVM bug since Log4Shell, but not for ease of exploitation. The overwritten bytes are a fixed pattern derived from the decoder, not attacker-chosen, and the Harness team did not reach code execution in their lab, only a crash after deliberately shrinking TLAB allocation buffers. A Lobsters thread dismissed the comparison, quoting an EPSS score of 0.00734, and Aikido had recorded no confirmed exploitation in the wild as of November 18. Calif, though, reports finding the flaw in about an hour and building a working exploit in two more, for under $200 in token costs, and is withholding per-product chains until patching catches up.
Four checks that find hidden runtimes
The dev.to post prescribes four audits. First, check /proc/cpuinfo for avx512f, avx512bw, avx512vbmi and bmi2 — all four flags must appear, though a clean dev box proves nothing if autoscaling can land the same image on newer silicon tomorrow. Second, sweep the filesystem for release files, which every JDK, JRE and jlink-built runtime has at its root; this surfaces vendor JREs under /opt, jpackage'd desktop tools, IDE runtimes and monitoring agents without executing anything. Third, inspect live processes by matching libjvm.so in /proc/*/maps rather than process names, then use jcmd to read the build number and VM flags. Fourth, for container images including shell-less distroless ones, export the filesystem with docker export and read the release file from the tarball; in Kubernetes, pull the image list from pod specs and repeat. Harness also published a checker script — exit code 2 signals exposure — though the post warns against piping it unread into a root shell.
One Temurin timing trap: Adoptium publishes binaries platform by platform as testing finishes, and Docker Hub images follow only once every binary for a version is out, so a rebuilt eclipse-temurin:17-jre image can silently pull the old JRE.
Stopgaps
Calif lists two flags as interim measures while still urging immediate upgrades: -XX:-UseBASE64Intrinsics forces the pure-Java decoder, the targeted fix at some performance cost on Base64-heavy workloads per Harness, and -XX:UseAVX=2, which Red Hat recommends, disables AVX-512 process-wide but will stop an ARM64 JVM from booting, so it must not ship in multi-arch images. Strict Base64 validation in front of the decoder acts as a backstop.
Why it matters
Java runtimes now proliferate far beyond the system package manager's reach: one per container image, one bundled per application, one per jlink build. Java4Shell is the rare bug whose remediation depends entirely on an inventory most teams do not have, which is why the dev.to post centers on discovery rather than the patch itself. The hardware gate compounds this — a fleet can look clean today and become exposed the moment a scheduler places a workload on AVX-512-capable silicon. Even if the exploitability debate resolves in defenders' favor, enumerating every JVM on disk and inside images is the durable lesson; the next runtime-level vulnerability will not wait for that inventory to be built.
- #java
- #security
- #containers
- #jvm
- #vulnerabilities