Zscaler 部落格

訂閱 Zscaler 部落格,第一時間獲取最新資訊!

Security Research

CVE-2026-114938: OpenJDK Base64 Decoder Vulnerable To Pre-Authentication RCE (Java4Shell)

image
RUBIN AZAD
November 18, 2026 - 7 分鐘閱讀
>CVE-2026-114938: OpenJDK Base64 Decoder Vulnerable To Pre-Authentication RCE (Java4Shell)

Introduction

On November 17, 2026, security researchers from Calif.io publicly disclosed Java4Shell (CVE-2026-114938, CVSS v4.0: 9.2 Critical), a severe heap memory-corruption vulnerability in OpenJDK’s hardware-accelerated (intrinsic) Base64 decoder on certain CPUs. The disclosure coincided with coordinated advisories from the OpenJDK Vulnerability Group and Oracle’s November 17, 2026 security patch release.

Because Base64 decoding is foundational to network protocols, this runtime-level vulnerability can be reached in enterprise applications that pass attacker-controlled input to the affected decoder. Calif.io demonstrated pre-authentication Remote Code Execution (RCE) in default configurations of Keycloak, Atlassian Jira, Atlassian Confluence, JFrog Artifactory, Jenkins, and PingFederate. Additionally, post-authentication RCE was achieved in Apache Kafka Broker, Apache Solr, Elasticsearch, Gerrit, and Netflix Spinnaker, as well as a client-side TLS exploit affecting Java clients connecting to a malicious server. 

Immediate patching for affected environments is strongly recommended, alongside other mitigation steps outlined in this advisory.  

Zscaler’s security team is actively analyzing this issue across both production and corporate environments. We will continue to update the Zscaler Trust post and this advisory as new information becomes available.

Recommendations

Organizations running Java workloads on x86-64 infrastructure should take the following actions:

  • Apply vendor patches immediately: Upgrade OpenJDK, Oracle Java SE, Oracle GraalVM, and all downstream OpenJDK distributions to the patched builds released on or after November 17, 2026 (such as Oracle Java SE 17.0.21.1, 8u513, 21.0.13.1, 25.0.5.1, and 27.0.1.1 security updates) and restart all running JVM processes and containers.
  • Apply JVM flag workarounds where patching is delayed: If an immediate JDK upgrade is not feasible, defenders can completely disable the vulnerable assembly code path by passing either of the following flags at JVM startup (e.g., via JAVA_TOOL_OPTIONS or JDK_JAVA_OPTIONS):
  • -XX:-UseBASE64Intrinsics (Recommended): Disables hardware-accelerated Base64 encoding and decoding intrinsics across all architectures, falling back to safe Java bytecode execution.
  • -XX:UseAVX=2 (x86-64 hosts only): Restricts HotSpot to AVX2 instructions, preventing the use of AVX-512 VBMI instructions. Warning: Use this flag only on x86-64 machines; OpenJDK on ARM64 does not recognize this flag and will fail to start.
  • After adjusting startup command lines and parameters, restart all running JVM processes and containers.
  • Check Host CPUs for AVX-512 VBMI Exposure: Because CVE-2026-114938 is hardware-dependent, administrators can determine if the CPU of a Linux system is capable of executing the vulnerable Base64 decoder by inspecting the /proc/cpuinfo file and determining if all four of the following flags are present: avx512f, avx512bw, avx512vbmi, and bmi2. If a system is not susceptible, the JVM patches should still be applied in accordance with local policy and best practices.
  • Eliminate public exposure: Place internet-facing Java applications (such as Keycloak, Jira, Confluence, Jenkins, Artifactory, and PingFederate) behind Zscaler Private Access (ZPA).
  • Enable reverse-proxy request buffering (Defense-in-Depth): Many remote heap grooming techniques rely on partial HTTP requests to prepare the heap for successful exploitation. Enabling full HTTP request-body buffering at the reverse proxy (e.g., Nginx proxy_request_buffering on;) disrupts multi-connection timing alignment—though note this is not a complete mitigation for single-request chains (such as Elasticsearch) or upgraded WebSocket tunnels.

Affected JDK Distributions And Versions

CVE-2026-114938 affects OpenJDK 17.0.6 and newer, as well as specific performance/patch builds across Oracle Java SE and downstream OpenJDK distributions as shown in the table below:

Vendor / Category

Affected Products & Versions

Remediation / Patch

OpenJDK (Upstream)

8.5.0 – 8.5.11

Apply OpenJDK 2026-11-17 advisory patches, or if patching is delayed, use -XX:-UseBASE64Intrinsics as a temporary workaround.

Oracle Java SE / Oracle GraalVM

8.4.0 – 8.4.26

Upgrade to the applicable Oracle November 17, 2026 patch (e.g., Java SE 17.0.21.1).

Downstream OpenJDK Distributions

Amazon Corretto, Eclipse Temurin, Microsoft Build of OpenJDK, Azul Zulu, BellSoft Liberica, SAP SapMachine, Red Hat OpenJDK, Oracle OpenJDK

Upgrade to a build that fixes CVE-2026-114938, or if patching is delayed, apply -XX:-UseBASE64Intrinsics as a temporary workaround.

Table 1: Affected JDK distributions and versions.

Hardware prerequisites

The vulnerable assembly routine is only emitted by the HotSpot C2 JIT compiler on x86-64 processors that support AVX-512 VBMI (avx512f, avx512bw, avx512vbmi, bmi2):

  • Vulnerable Intel server CPUs: Intel Xeon Ice Lake, Sapphire Rapids, Emerald Rapids, and Granite Rapids
  • Vulnerable AMD server and desktop CPUs: AMD EPYC and Ryzen Zen 4 or newer (e.g., EPYC Genoa / Bergamo / Turin, Ryzen 7000/9000 series)
  • Not affected: ARM64 processors (e.g., AWS Graviton, Apple Silicon, Ampere Altra) and older x86-64 CPUs lacking AVX-512 VBMI.

Technical Analysis

How it works

While Java is traditionally regarded as a memory-safe language that abstracts raw pointer arithmetic away from developers, the OpenJDK HotSpot Virtual Machine itself relies heavily on C, C++, and hand-rolled assembly intrinsics to maximize throughput on modern hardware. Specifically, OpenJDK’s C2 Just-In-Time (JIT) compiler includes an optimized x86-64 Base64 decoding routine leveraging AVX-512 VBMI vector instructions. Due to an arithmetic underflow when calculating the store mask for improperly padded Base64 tails, malformed Base64 input can be used to trigger a 62-byte heap out-of-bounds write.

Base64 decoding sits at the foundation of many network protocols including HTTP authentication, cookies, authorization tokens, TLS certificate parsing, REST API payloads, and file uploads. Thus, the attack surface for this vulnerability is significant. The figure below shows how malformed input can lead to heap corruption.

Java4Shell

Figure 1: CVE-2026-114938: Java4Shell Exploit Workflow

Conclusion

CVE-2026-114938 (Java4Shell) demonstrates that even memory-safe managed runtimes like Java are susceptible to classic memory corruption when low-level assembly intrinsics miscalculate bounds. Organizations running affected Java builds on x86-64 processors with the required instruction support should patch immediately . Where immediate patching is not possible, restart affected JVMs with -XX:-UseBASE64Intrinsics to temporarily disable the vulnerable code path.

Zscaler Coverage

Zscaler has deployed the following protections for CVE-2026-114938:

Zscaler Deception

Zscaler Deception customers can deploy Private Threat Intelligence decoys for the following applications to detect Java4Shell exploitation attempts and generate high-fidelity alerts:

  • Apache Solr
  • Jenkins
  • Atlassian Confluence
  • Atlassian Jira
  • Elasticsearch

Customers can configure an orchestration rule to add any IP addresses interacting with these decoys to a firewall block list.

How Zscaler Can Help

The following table outlines typical attack stages for CVE-2026-114938 and the mitigations recommended by Zscaler:

Attack Stage

Recommended Mitigation

Minimize the external attack surface

Prevent compromise

  • Inspect inbound and outbound traffic with ZPA and Zscaler Internet Access (ZIA) to detect and block malformed Base64 padding tails, protect Java clients against the client-side TLS exploit vector by routing egress traffic through ZIA with full SSL/TLS inspection.

Prevent lateral threat movement

  • Use ZPA to enforce least-privilege user-to-app and app-to-app microsegmentation for crown-jewel workloads, deploy Zscaler Deception decoys to detect lateral movement.
  • Use Zscaler Microsegmentation to restrict workload-to-workload traffic and limit lateral movement if a Java workload is compromised.

Prevent data loss

  • Inspect outbound responses and egress traffic across all channels with Zscaler Data Loss Prevention (DLP) to detect and block JavaBleed heap memory leaks containing private keys, credentials, or PII.

 

Indicators Of Compromise (IOCs)

Network-level indicators

  • JIT warm-up floods: High volumes of rapid, repetitive requests containing valid Base64 fields targeting endpoints that invoke java.util.Base64.Decoder.
  • Connection parking (remote heap grooming): Multiple concurrent TCP/HTTP connections from the same source IP delivering partial HTTP request bodies, pausing awaiting TCP ACK, and releasing the final bytes simultaneously.
  • Malformed Base64 tail triggers: Base64 strings whose length is a multiple of 64 bytes (64 or 128 characters) immediately followed by an invalid padded tail (=, ==, X=, X==, XY=, ===), such as AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA= (64 bytes + =) or 128 × "A" + "=".
  • JavaBleed response anomalies: Abnormally large HTTP responses returned immediately after a spike in IllegalArgumentException errors.

Host-level indicators

  • Spikes in java.lang.IllegalArgumentException: High rates of IllegalArgumentException: Illegal base64 character or invalid padding exceptions in application logs.
  • JVM crash logs (hs_err_pid*.log): Repeated HotSpot SIGSEGV crashes in G1 GC routines or StubRoutines::base64_decodeBlock.
  • Suspicious JVM child processes: Java process spawning unexpected OS shells or reconnaissance commands.
form submtited
感謝您的閱讀

這篇文章對您有幫助嗎?

免責聲明:本部落格文章由 Zscaler 製作,僅供資訊參考,並以「現狀」提供,不保證其準確性、完整性或可靠性。Zscaler 對任何錯誤或遺漏,或基於所提供的資訊而採取的任何行動不承擔任何責任。本部落格文章中連結的任何第三方網站或資源僅為方便起見而提供,Zscaler 對其內容或做法概不負責。所有內容如有變更,恕不另行通知。存取此部落格,即表示您同意這些條款,並承認您有責任自行驗證並根據您的需求適當使用資訊。

訂閱 Zscaler 部落格,第一時間獲取最新資訊!

提交此表格即表示您同意我們的隱私權政策。