Repository navigation
Conversation
CVE-2026-114938 is a critical heap-overflow RCE in the JDK's AVX-512 Base64 decoder intrinsic, disclosed 2026-11-17. It affects OpenJDK 17.0.6+, 21, 25 and 27 (not 8 or 11) on x86-64 hosts with AVX-512 VBMI support. Researchers demonstrated post-auth RCE against Solr; both BasicAuthPlugin and PKIAuthenticationPlugin decode request-supplied Base64 input directly. No JDK vendor (Temurin, Zulu, Oracle's own OSS builds, Debian) has shipped a GA patched build for the 17/21/25 lines yet, so there is no dependency bump that fixes this. This documents the universally recommended interim mitigation, -XX:-UseBASE64Intrinsics, in solr.in.sh/solr.in.cmd as an opt-in commented-out setting, following the exact same pattern already used here for CVE-2021-44228 (log4j2.formatMsgNoLookups). The flag is a generic HotSpot product flag (globals.hpp) honored on every architecture, so enabling it is a no-op on platforms this CVE doesn't affect. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
|
On the dev community meeting today it was suggested that an announcement can be made in the "Security News" section of the page and that JDK is not a dependency but a requirement, like having a machine that Solr can run on, and since we can't give security advisories for all the different Operating Systems, it's not viable to do so for JDK either, and therefore an announcement could go in Security News. @epugh was mentioned as the one who should make this change. |
Unlike bin/solr.in.sh, where the equivalent setting is left as an opt-in comment, enable -XX:-UseBASE64Intrinsics by default in the Docker image's /etc/default/solr.in.sh. Users who just pull the published image don't necessarily read solr.in.sh, and this is a shipped artifact the project already controls end to end, unlike third-party JDK builds. Verified via createDockerfileLocal/createDockerfileOfficialFull/ createDockerfileOfficialSlim that the rendered Dockerfiles pick this up correctly. Could not do a live `docker build` in this environment (no local Docker daemon). Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
That's understandable from a Solr application point of view, but you also maintain and distribute a docker image that currently contains a vulnerable version of the JDK. When can we expect an advisory and update for this?
|
there is a nuance there that is easily overlooked: the "apache solr project" does not build and publish the docker images that are found at https://hub.docker.com/_/solr (the text you quoted is regrettably misleading.) The Dockerfile used to build and publish those docker images IS "built and maintained in the official Solr repo github.com/apache/solr" ... but the docker.io "docker-library" project uses a pointer to that Dockerfile -- provided by the apache solr project and other OSS projects like ours -- to build all "official" docker images , and regularly rebuilds them when their base image changes...
So for example: soon as there is a new official |
dsmiley
left a comment
There was a problem hiding this comment.
+1 to the PR and thanks for contributing
I don't think there's any urgency as it's something a user can do from their end with SOLR_OPTS env var or --jvm-opts CLI option. If a user is proactive/alert to Java4Shell like you, they can do this now. I expect Docker Inc & Eclipse Temurin (JDK) to release an update much faster than we can do so. You can also expect an update to our website shortly to advise users on what I said and more.
|
Automated PR readiness judgements (LLM-generated; may be wrong; not updated on later pushes).
|
|
The PR readiness judgements half sucked this time :-( Changelog -- great feedback; JIRA & upgrade notes -- no. |
What
Two changes for CVE-2026-114938 ("Java4Shell"), a critical heap-overflow RCE in the JDK's AVX-512 Base64 decoder intrinsic:
solr.in.sh/solr.in.cmd: documents-XX:-UseBASE64Intrinsicsas an opt-in, commented-outSOLR_OPTSsetting, mirroring the existing pattern this repo already uses for CVE-2021-44228 (thelog4j2.formatMsgNoLookupsline a few lines above it).solr/docker/templates/Dockerfile.body.template: enables the same flag by default (not commented out) in the Docker image's/etc/default/solr.in.sh, since someone who just pulls the published image doesn't necessarily readsolr.in.shfirst. I confirmed viaapache/solr-docker's own README that its Dockerfile is generated from this exact template at release time, and that repo explicitly refuses manual edits — so this is the actual source of truth for the published image's behavior.Why
Disclosed 2026-11-17. Affects OpenJDK 17.0.6+, 21, 25 and 27 (not 8 or 11) on x86-64 hosts with AVX-512
VBMI (Intel Ice Lake+ Xeon, AMD Zen 4+ EPYC/Ryzen). The disclosing researchers (Calif) demonstrated
post-auth RCE against Apache Solr specifically. Both
BasicAuthPluginandPKIAuthenticationPlugincall
Base64.getDecoder().decode()directly on request-supplied input, which is real exposed surfaceregardless of exactly which code path the researchers used.
As of this PR, no JDK vendor has a GA build with the fix for the 17/21/25 lines. I went and checked the
actual OpenJDK source history: the real fix (
8393212, a 4-line guard inStubGenerator::generate_base64_decodeBlock()) already landed injdk17u-dev/jdk21u-dev/jdk25u-devon 2026-11-12, and the GA source tags (
jdk-17.0.21.1-ga,jdk-21.0.13.1-ga,jdk-25.0.5.1-ga)were cut 2026-11-15 — but no vendor (Temurin, Zulu, Oracle's own OSS builds, Debian) has published a GA
binary build yet. Checked the Adoptium API directly: Temurin's current
25.0.5.1tag is stillEA/beta (
25.0.5.1-beta+1-ea), and it's actually built from an earlier pre-fix commit than the one thatmatters, despite sharing the version string. So there's no dependency/base-image bump that resolves this
today, and operators need a flag-based mitigation in the meantime.
-XX:-UseBASE64Intrinsicsis a generic HotSpot product flag (globals.hpp, checked in the JDK source),honored identically across x86, aarch64, ppc and riscv builds — so turning it off is a safe no-op on
platforms this CVE doesn't affect, not just an x86-64-specific workaround.
Why default-on for Docker but opt-in for solr.in.sh
The tarball/
solr.in.shaudience reads that file to configure their deployment, same as the existingLog4Shell precedent. The Docker image audience often doesn't touch
solr.in.shat all — they just pullsolr:10.1.0and run it. Given the severity (critical, post-auth RCE demonstrated against Solr) and thatthe flag is a safe no-op where it doesn't apply, defaulting it on in the one artifact the project fully
controls end-to-end seemed like the right asymmetry. Open to the maintainers disagreeing on this point —
flagging it explicitly rather than assuming it's uncontroversial.
How to verify
./gradlew :solr:docker:createDockerfileLocal :solr:docker:createDockerfileOfficialFull :solr:docker:createDockerfileOfficialSlim— all three generated Dockerfiles pick up the new line correctly.docker build/docker runin this environment (no local Docker daemon) — would appreciate a maintainer or CI doing that smoke test before this merges.Notes for reviewers
project=SOLR AND text~"Java4Shell", no hits), so this PR doesn't carry aSOLR-NNNNNprefix. Happy to file one ifa committer wants it tracked that way.
project wants to handle a JDK-level CVE with no code fix available, especially the default-on-for-Docker
question above.
type: security), following the project's documentedwriteChangelogPrworkflow, styled afterSOLR-17888's Tika-CVE mitigation entry.Co-Authored-By: Claude Sonnet 5 noreply@anthropic.com